Auto merge of #91003 - psumbera:sparc64-abi, r=nagisa

fix sparc64 ABI for aggregates with floating point members

Fixes #86163

compiler/rustc_codegen_cranelift/src/abi/pass_mode.rs

@@ -71,7 +71,7 @@ fn cast_target_to_abi_params(cast: CastTarget) -> SmallVec<[AbiParam; 2]> {
         .prefix
         .iter()
         .flatten()
-        .map(|&kind| reg_to_abi_param(Reg { kind, size: cast.prefix_chunk_size }))
+        .map(|&reg| reg_to_abi_param(reg))
         .chain((0..rest_count).map(|_| reg_to_abi_param(cast.rest.unit)))
         .collect::<SmallVec<_>>();
 

compiler/rustc_codegen_gcc/src/abi.rs

@@ -48,8 +48,8 @@ impl GccType for CastTarget {
         let mut args: Vec<_> = self
             .prefix
             .iter()
-            .flat_map(|option_kind| {
-                option_kind.map(|kind| Reg { kind, size: self.prefix_chunk_size }.gcc_type(cx))
+            .flat_map(|option_reg| {
+                option_reg.map(|reg| reg.gcc_type(cx))
             })
             .chain((0..rest_count).map(|_| rest_gcc_unit))
             .collect();

compiler/rustc_codegen_llvm/src/abi.rs

@@ -181,9 +181,7 @@ impl LlvmType for CastTarget {
         let mut args: Vec<_> = self
             .prefix
             .iter()
-            .flat_map(|option_kind| {
-                option_kind.map(|kind| Reg { kind, size: self.prefix_chunk_size }.llvm_type(cx))
-            })
+            .flat_map(|option_reg| option_reg.map(|reg| reg.llvm_type(cx)))
             .chain((0..rest_count).map(|_| rest_ll_unit))
             .collect();
 
@@ -466,6 +464,9 @@ impl<'tcx> FnAbiLlvmExt<'tcx> for FnAbi<'tcx, Ty<'tcx>> {
                     );
                 }
             }
+            PassMode::Cast(cast) => {
+                cast.attrs.apply_attrs_to_llfn(llvm::AttributePlace::ReturnValue, cx, llfn);
+            }
             _ => {}
         }
         for arg in &self.args {
@@ -497,8 +498,8 @@ impl<'tcx> FnAbiLlvmExt<'tcx> for FnAbi<'tcx, Ty<'tcx>> {
                     apply(a);
                     apply(b);
                 }
-                PassMode::Cast(_) => {
-                    apply(&ArgAttributes::new());
+                PassMode::Cast(cast) => {
+                    apply(&cast.attrs);
                 }
             }
         }
@@ -533,6 +534,13 @@ impl<'tcx> FnAbiLlvmExt<'tcx> for FnAbi<'tcx, Ty<'tcx>> {
                     );
                 }
             }
+            PassMode::Cast(cast) => {
+                cast.attrs.apply_attrs_to_callsite(
+                    llvm::AttributePlace::ReturnValue,
+                    &bx.cx,
+                    callsite,
+                );
+            }
             _ => {}
         }
         if let abi::Abi::Scalar(scalar) = self.ret.layout.abi {
@@ -577,8 +585,8 @@ impl<'tcx> FnAbiLlvmExt<'tcx> for FnAbi<'tcx, Ty<'tcx>> {
                     apply(bx.cx, a);
                     apply(bx.cx, b);
                 }
-                PassMode::Cast(_) => {
-                    apply(bx.cx, &ArgAttributes::new());
+                PassMode::Cast(cast) => {
+                    apply(bx.cx, &cast.attrs);
                 }
             }
         }

compiler/rustc_target/src/abi/call/mips64.rs

@@ -1,4 +1,6 @@
-use crate::abi::call::{ArgAbi, ArgExtension, CastTarget, FnAbi, PassMode, Reg, RegKind, Uniform};
+use crate::abi::call::{
+    ArgAbi, ArgAttribute, ArgAttributes, ArgExtension, CastTarget, FnAbi, PassMode, Reg, Uniform,
+};
 use crate::abi::{self, HasDataLayout, Size, TyAbiInterface};
 
 fn extend_integer_width_mips<Ty>(arg: &mut ArgAbi<'_, Ty>, bits: u64) {
@@ -115,7 +117,7 @@ where
                             for _ in 0..((offset - last_offset).bits() / 64)
                                 .min((prefix.len() - prefix_index) as u64)
                             {
-                                prefix[prefix_index] = Some(RegKind::Integer);
+                                prefix[prefix_index] = Some(Reg::i64());
                                 prefix_index += 1;
                             }
 
@@ -123,7 +125,7 @@ where
                                 break;
                             }
 
-                            prefix[prefix_index] = Some(RegKind::Float);
+                            prefix[prefix_index] = Some(Reg::f64());
                             prefix_index += 1;
                             last_offset = offset + Reg::f64().size;
                         }
@@ -137,8 +139,13 @@ where
     let rest_size = size - Size::from_bytes(8) * prefix_index as u64;
     arg.cast_to(CastTarget {
         prefix,
-        prefix_chunk_size: Size::from_bytes(8),
         rest: Uniform { unit: Reg::i64(), total: rest_size },
+        attrs: ArgAttributes {
+            regular: ArgAttribute::default(),
+            arg_ext: ArgExtension::None,
+            pointee_size: Size::ZERO,
+            pointee_align: None,
+        },
     });
 }
 

compiler/rustc_target/src/abi/call/mod.rs

@@ -214,9 +214,9 @@ impl Uniform {
 
 #[derive(Clone, Copy, PartialEq, Eq, Hash, Debug, HashStable_Generic)]
 pub struct CastTarget {
-    pub prefix: [Option<RegKind>; 8],
-    pub prefix_chunk_size: Size,
+    pub prefix: [Option<Reg>; 8],
     pub rest: Uniform,
+    pub attrs: ArgAttributes,
 }
 
 impl From<Reg> for CastTarget {
@@ -227,29 +227,48 @@ impl From<Reg> for CastTarget {
 
 impl From<Uniform> for CastTarget {
     fn from(uniform: Uniform) -> CastTarget {
-        CastTarget { prefix: [None; 8], prefix_chunk_size: Size::ZERO, rest: uniform }
+        CastTarget {
+            prefix: [None; 8],
+            rest: uniform,
+            attrs: ArgAttributes {
+                regular: ArgAttribute::default(),
+                arg_ext: ArgExtension::None,
+                pointee_size: Size::ZERO,
+                pointee_align: None,
+            },
+        }
     }
 }
 
 impl CastTarget {
     pub fn pair(a: Reg, b: Reg) -> CastTarget {
         CastTarget {
-            prefix: [Some(a.kind), None, None, None, None, None, None, None],
-            prefix_chunk_size: a.size,
+            prefix: [Some(a), None, None, None, None, None, None, None],
             rest: Uniform::from(b),
+            attrs: ArgAttributes {
+                regular: ArgAttribute::default(),
+                arg_ext: ArgExtension::None,
+                pointee_size: Size::ZERO,
+                pointee_align: None,
+            },
         }
     }
 
-    pub fn size<C: HasDataLayout>(&self, cx: &C) -> Size {
-        (self.prefix_chunk_size * self.prefix.iter().filter(|x| x.is_some()).count() as u64)
-            .align_to(self.rest.align(cx))
-            + self.rest.total
+    pub fn size<C: HasDataLayout>(&self, _cx: &C) -> Size {
+        let mut size = self.rest.total;
+        for i in 0..self.prefix.iter().count() {
+            match self.prefix[i] {
+                Some(v) => size += Size { raw: v.size.bytes() },
+                None => {}
+            }
+        }
+        return size;
     }
 
     pub fn align<C: HasDataLayout>(&self, cx: &C) -> Align {
         self.prefix
             .iter()
-            .filter_map(|x| x.map(|kind| Reg { kind, size: self.prefix_chunk_size }.align(cx)))
+            .filter_map(|x| x.map(|reg| reg.align(cx)))
             .fold(cx.data_layout().aggregate_align.abi.max(self.rest.align(cx)), |acc, align| {
                 acc.max(align)
             })

compiler/rustc_target/src/abi/call/sparc64.rs

@@ -1,7 +1,9 @@
 // FIXME: This needs an audit for correctness and completeness.
 
-use crate::abi::call::{ArgAbi, FnAbi, Reg, RegKind, Uniform};
-use crate::abi::{HasDataLayout, TyAbiInterface};
+use crate::abi::call::{
+    ArgAbi, ArgAttribute, ArgAttributes, ArgExtension, CastTarget, FnAbi, Reg, RegKind, Uniform,
+};
+use crate::abi::{self, HasDataLayout, Size, TyAbiInterface};
 
 fn is_homogeneous_aggregate<'a, Ty, C>(cx: &C, arg: &mut ArgAbi<'a, Ty>) -> Option<Uniform>
 where
@@ -16,7 +18,7 @@ where
 
         let valid_unit = match unit.kind {
             RegKind::Integer => false,
-            RegKind::Float => true,
+            RegKind::Float => false,
             RegKind::Vector => arg.layout.size.bits() == 128,
         };
 
@@ -24,33 +26,7 @@ where
     })
 }
 
-fn classify_ret<'a, Ty, C>(cx: &C, ret: &mut ArgAbi<'a, Ty>)
-where
-    Ty: TyAbiInterface<'a, C> + Copy,
-    C: HasDataLayout,
-{
-    if !ret.layout.is_aggregate() {
-        ret.extend_integer_width_to(64);
-        return;
-    }
-
-    if let Some(uniform) = is_homogeneous_aggregate(cx, ret) {
-        ret.cast_to(uniform);
-        return;
-    }
-    let size = ret.layout.size;
-    let bits = size.bits();
-    if bits <= 256 {
-        let unit = Reg::i64();
-        ret.cast_to(Uniform { unit, total: size });
-        return;
-    }
-
-    // don't return aggregates in registers
-    ret.make_indirect();
-}
-
-fn classify_arg<'a, Ty, C>(cx: &C, arg: &mut ArgAbi<'a, Ty>)
+fn classify_arg<'a, Ty, C>(cx: &C, arg: &mut ArgAbi<'a, Ty>, in_registers_max: Size)
 where
     Ty: TyAbiInterface<'a, C> + Copy,
     C: HasDataLayout,
@@ -60,13 +36,97 @@ where
         return;
     }
 
+    // This doesn't intentionally handle structures with floats which needs
+    // special care below.
     if let Some(uniform) = is_homogeneous_aggregate(cx, arg) {
         arg.cast_to(uniform);
         return;
     }
 
+    if let abi::FieldsShape::Arbitrary { .. } = arg.layout.fields {
+        let dl = cx.data_layout();
+        let size = arg.layout.size;
+        let mut prefix = [None; 8];
+        let mut prefix_index = 0;
+        let mut last_offset = Size::ZERO;
+        let mut has_float = false;
+        let mut arg_attribute = ArgAttribute::default();
+
+        for i in 0..arg.layout.fields.count() {
+            let field = arg.layout.field(cx, i);
+            let offset = arg.layout.fields.offset(i);
+
+            if let abi::Abi::Scalar(scalar) = &field.abi {
+                if scalar.value == abi::F32 || scalar.value == abi::F64 {
+                    has_float = true;
+
+                    if !last_offset.is_aligned(dl.f64_align.abi) && last_offset < offset {
+                        if prefix_index == prefix.len() {
+                            break;
+                        }
+                        prefix[prefix_index] = Some(Reg::i32());
+                        prefix_index += 1;
+                        last_offset = last_offset + Reg::i32().size;
+                    }
+
+                    for _ in 0..((offset - last_offset).bits() / 64)
+                        .min((prefix.len() - prefix_index) as u64)
+                    {
+                        prefix[prefix_index] = Some(Reg::i64());
+                        prefix_index += 1;
+                        last_offset = last_offset + Reg::i64().size;
+                    }
+
+                    if last_offset < offset {
+                        if prefix_index == prefix.len() {
+                            break;
+                        }
+                        prefix[prefix_index] = Some(Reg::i32());
+                        prefix_index += 1;
+                        last_offset = last_offset + Reg::i32().size;
+                    }
+
+                    if prefix_index == prefix.len() {
+                        break;
+                    }
+
+                    if scalar.value == abi::F32 {
+                        arg_attribute = ArgAttribute::InReg;
+                        prefix[prefix_index] = Some(Reg::f32());
+                        last_offset = offset + Reg::f32().size;
+                    } else {
+                        prefix[prefix_index] = Some(Reg::f64());
+                        last_offset = offset + Reg::f64().size;
+                    }
+                    prefix_index += 1;
+                }
+            }
+        }
+
+        if has_float && arg.layout.size <= in_registers_max {
+            let mut rest_size = size - last_offset;
+
+            if (rest_size.raw % 8) != 0 && prefix_index < prefix.len() {
+                prefix[prefix_index] = Some(Reg::i32());
+                rest_size = rest_size - Reg::i32().size;
+            }
+
+            arg.cast_to(CastTarget {
+                prefix,
+                rest: Uniform { unit: Reg::i64(), total: rest_size },
+                attrs: ArgAttributes {
+                    regular: arg_attribute,
+                    arg_ext: ArgExtension::None,
+                    pointee_size: Size::ZERO,
+                    pointee_align: None,
+                },
+            });
+            return;
+        }
+    }
+
     let total = arg.layout.size;
-    if total.bits() > 128 {
+    if total > in_registers_max {
         arg.make_indirect();
         return;
     }
@@ -80,13 +140,13 @@ where
     C: HasDataLayout,
 {
     if !fn_abi.ret.is_ignore() {
-        classify_ret(cx, &mut fn_abi.ret);
+        classify_arg(cx, &mut fn_abi.ret, Size { raw: 32 });
     }
 
     for arg in &mut fn_abi.args {
         if arg.is_ignore() {
             continue;
         }
-        classify_arg(cx, arg);
+        classify_arg(cx, arg, Size { raw: 16 });
     }
 }

src/test/assembly/sparc-struct-abi.rs

@@ -0,0 +1,64 @@
+// Test SPARC64 ABI
+// - float structure members are passes in floating point registers
+// (#86163)
+
+// assembly-output: emit-asm
+// needs-llvm-components: sparc
+// compile-flags: --target=sparcv9-sun-solaris -Copt-level=3
+#![crate_type = "lib"]
+#![feature(no_core, lang_items)]
+#![no_core]
+
+#[lang = "sized"]
+pub trait Sized {}
+#[lang = "copy"]
+pub trait Copy {}
+
+#[repr(C)]
+pub struct Franta {
+    a: f32,
+    b: f32,
+    c: f32,
+    d: f32,
+}
+
+// NB: due to delay slots the `ld` following the call is actually executed before the call.
+#[no_mangle]
+pub unsafe extern "C" fn callee(arg: Franta) {
+    // CHECK-LABEL: callee:
+    // CHECK: st %f3, [[PLACE_D:.*]]
+    // CHECK: st %f2, [[PLACE_C:.*]]
+    // CHECK: st %f1, [[PLACE_B:.*]]
+    // CHECK: st %f0, [[PLACE_A:.*]]
+    // CHECK: call tst_use
+    // CHECK-NEXT: ld [[PLACE_A]], %f1
+    // CHECK: call tst_use
+    // CHECK-NEXT: ld [[PLACE_B]], %f1
+    // CHECK: call tst_use
+    // CHECK-NEXT: ld [[PLACE_C]], %f1
+    // CHECK: call tst_use
+    // CHECK-NEXT: ld [[PLACE_D]], %f1
+    clobber();
+    tst_use(arg.a);
+    tst_use(arg.b);
+    tst_use(arg.c);
+    tst_use(arg.d);
+}
+
+extern "C" {
+    fn opaque_callee(arg: Franta, intarg: i32);
+    fn tst_use(arg: f32);
+    fn clobber();
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn caller() {
+    // CHECK-LABEL: caller:
+    // CHECK: ld [{{.*}}], %f0
+    // CHECK: ld [{{.*}}], %f1
+    // CHECK: ld [{{.*}}], %f2
+    // CHECK: ld [{{.*}}], %f3
+    // CHECK: call opaque_callee
+    // CHECK: mov     3, %o2
+    opaque_callee(Franta { a: 1.0, b: 2.0, c: 3.0, d: 4.0 }, 3);
+}

src/test/codegen/sparc-struct-abi.rs

@@ -1,5 +1,5 @@
 // Checks that we correctly codegen extern "C" functions returning structs.
-// See issue #52638.
+// See issues #52638 and #86163.
 
 // compile-flags: -O --target=sparc64-unknown-linux-gnu --crate-type=rlib
 // needs-llvm-components: sparc
@@ -25,3 +25,59 @@ pub struct Bool {
 pub extern "C" fn structbool() -> Bool {
     Bool { b: true }
 }
+
+
+#[repr(C)]
+pub struct BoolFloat {
+    b: bool,
+    f: f32,
+}
+
+// CHECK: define inreg { i32, float } @structboolfloat()
+// CHECK-NEXT: start:
+// CHECK-NEXT: ret { i32, float } { i32 16777216, float 0x40091EB860000000 }
+#[no_mangle]
+pub extern "C" fn structboolfloat() -> BoolFloat {
+    BoolFloat { b: true, f: 3.14 }
+}
+
+// CHECK: define void @structboolfloat_input({ i32, float } inreg %0)
+// CHECK-NEXT: start:
+#[no_mangle]
+pub extern "C" fn structboolfloat_input(a: BoolFloat) { }
+
+
+#[repr(C)]
+pub struct ShortDouble {
+    s: i16,
+    d: f64,
+}
+
+// CHECK: define { i64, double } @structshortdouble()
+// CHECK-NEXT: start:
+// CHECK-NEXT: ret { i64, double } { i64 34621422135410688, double 3.140000e+00 }
+#[no_mangle]
+pub extern "C" fn structshortdouble() -> ShortDouble {
+    ShortDouble { s: 123, d: 3.14 }
+}
+
+// CHECK: define void @structshortdouble_input({ i64, double } %0)
+// CHECK-NEXT: start:
+#[no_mangle]
+pub extern "C" fn structshortdouble_input(a: ShortDouble) { }
+
+
+#[repr(C)]
+pub struct FloatLongFloat {
+    f: f32,
+    i: i64,
+    g: f32,
+}
+
+// CHECK: define inreg { float, i32, i64, float, i32 } @structfloatlongfloat()
+// CHECK-NEXT: start:
+// CHECK-NEXT: ret { float, i32, i64, float, i32 } { float 0x3FB99999A0000000, i32 undef, i64 123, float 0x40091EB860000000, i32 undef }
+#[no_mangle]
+pub extern "C" fn structfloatlongfloat() -> FloatLongFloat {
+    FloatLongFloat { f: 0.1, i: 123, g: 3.14 }
+}