Refactor powerpc64 call ABI handling

compiler/rustc_target/src/abi/call/powerpc64.rs

@@ -41,64 +41,23 @@ where
     })
 }
 
-fn classify_ret<'a, Ty, C>(cx: &C, ret: &mut ArgAbi<'a, Ty>, abi: ABI)
+fn classify<'a, Ty, C>(cx: &C, arg: &mut ArgAbi<'a, Ty>, abi: ABI, is_ret: bool)
 where
     Ty: TyAbiInterface<'a, C> + Copy,
     C: HasDataLayout,
 {
-    if !ret.layout.is_sized() {
+    if arg.is_ignore() || !arg.layout.is_sized() {
         // Not touching this...
         return;
     }
-    if !ret.layout.is_aggregate() {
-        ret.extend_integer_width_to(64);
+    if !arg.layout.is_aggregate() {
+        arg.extend_integer_width_to(64);
         return;
     }
 
     // The ELFv1 ABI doesn't return aggregates in registers
-    if abi == ELFv1 {
-        ret.make_indirect();
-        return;
-    }
-
-    if let Some(uniform) = is_homogeneous_aggregate(cx, ret, abi) {
-        ret.cast_to(uniform);
-        return;
-    }
-
-    let size = ret.layout.size;
-    let bits = size.bits();
-    if bits <= 128 {
-        let unit = if cx.data_layout().endian == Endian::Big {
-            Reg { kind: RegKind::Integer, size }
-        } else if bits <= 8 {
-            Reg::i8()
-        } else if bits <= 16 {
-            Reg::i16()
-        } else if bits <= 32 {
-            Reg::i32()
-        } else {
-            Reg::i64()
-        };
-
-        ret.cast_to(Uniform::new(unit, size));
-        return;
-    }
-
-    ret.make_indirect();
-}
-
-fn classify_arg<'a, Ty, C>(cx: &C, arg: &mut ArgAbi<'a, Ty>, abi: ABI)
-where
-    Ty: TyAbiInterface<'a, C> + Copy,
-    C: HasDataLayout,
-{
-    if !arg.layout.is_sized() {
-        // Not touching this...
-        return;
-    }
-    if !arg.layout.is_aggregate() {
-        arg.extend_integer_width_to(64);
+    if is_ret && abi == ELFv1 {
+        arg.make_indirect();
         return;
     }
 
@@ -108,7 +67,10 @@ where
     }
 
     let size = arg.layout.size;
-    if size.bits() <= 64 {
+    if is_ret && size.bits() > 128 {
+        // Non-homogeneous aggregates larger than two doublewords are returned indirectly.
+        arg.make_indirect();
+    } else if size.bits() <= 64 {
         // Aggregates smaller than a doubleword should appear in
         // the least-significant bits of the parameter doubleword.
         arg.cast_to(Reg { kind: RegKind::Integer, size })
@@ -138,14 +100,9 @@ where
         }
     };
 
-    if !fn_abi.ret.is_ignore() {
-        classify_ret(cx, &mut fn_abi.ret, abi);
-    }
+    classify(cx, &mut fn_abi.ret, abi, true);
 
     for arg in fn_abi.args.iter_mut() {
-        if arg.is_ignore() {
-            continue;
-        }
-        classify_arg(cx, arg, abi);
+        classify(cx, arg, abi, false);
     }
 }

tests/assembly/powerpc64-struct-abi.rs

@@ -0,0 +1,132 @@
+//@ revisions: elfv1-be elfv2-be elfv2-le
+//@ assembly-output: emit-asm
+//@ compile-flags: -O
+//@[elfv1-be] compile-flags: --target powerpc64-unknown-linux-gnu
+//@[elfv1-be] needs-llvm-components: powerpc
+//@[elfv2-be] compile-flags: --target powerpc64-unknown-linux-musl
+//@[elfv2-be] needs-llvm-components: powerpc
+//@[elfv2-le] compile-flags: --target powerpc64le-unknown-linux-gnu
+//@[elfv2-le] needs-llvm-components: powerpc
+//@[elfv1-be] filecheck-flags: --check-prefix be
+//@[elfv2-be] filecheck-flags: --check-prefix be
+
+#![feature(no_core, lang_items)]
+#![no_std]
+#![no_core]
+#![crate_type = "lib"]
+
+#[lang = "sized"]
+trait Sized {}
+
+#[lang = "copy"]
+trait Copy {}
+
+#[lang = "freeze"]
+trait Freeze {}
+
+#[lang = "unpin"]
+trait Unpin {}
+
+impl Copy for u8 {}
+impl Copy for u16 {}
+impl Copy for u32 {}
+impl Copy for FiveU32s {}
+impl Copy for FiveU16s {}
+impl Copy for ThreeU8s {}
+
+#[repr(C)]
+struct FiveU32s(u32, u32, u32, u32, u32);
+
+#[repr(C)]
+struct FiveU16s(u16, u16, u16, u16, u16);
+
+#[repr(C)]
+struct ThreeU8s(u8, u8, u8);
+
+// CHECK-LABEL: read_large
+// be: lwz [[REG1:.*]], 16(4)
+// be-NEXT: stw [[REG1]], 16(3)
+// be-NEXT: ld [[REG2:.*]], 8(4)
+// be-NEXT: ld [[REG3:.*]], 0(4)
+// be-NEXT: std [[REG2]], 8(3)
+// be-NEXT: std [[REG3]], 0(3)
+// elfv2-le: lxvd2x [[REG1:.*]], 0, 4
+// elfv2-le-NEXT: lwz [[REG2:.*]], 16(4)
+// elfv2-le-NEXT: stw [[REG2]], 16(3)
+// elfv2-le-NEXT: stxvd2x [[REG1]], 0, 3
+// CHECK-NEXT: blr
+#[no_mangle]
+extern "C" fn read_large(x: &FiveU32s) -> FiveU32s {
+    *x
+}
+
+// CHECK-LABEL: read_medium
+// elfv1-be: lhz [[REG1:.*]], 8(4)
+// elfv1-be-NEXT: ld [[REG2:.*]], 0(4)
+// elfv1-be-NEXT: sth [[REG1]], 8(3)
+// elfv1-be-NEXT: std [[REG2]], 0(3)
+// elfv2-be: lhz [[REG1:.*]], 8(3)
+// elfv2-be-NEXT: ld 3, 0(3)
+// elfv2-be-NEXT: sldi 4, [[REG1]], 48
+// elfv2-le: ld [[REG1:.*]], 0(3)
+// elfv2-le-NEXT: lhz 4, 8(3)
+// elfv2-le-NEXT: mr 3, [[REG1]]
+// CHECK-NEXT: blr
+#[no_mangle]
+extern "C" fn read_medium(x: &FiveU16s) -> FiveU16s {
+    *x
+}
+
+// CHECK-LABEL: read_small
+// elfv1-be: lbz [[REG1:.*]], 2(4)
+// elfv1-be-NEXT: lhz [[REG2:.*]], 0(4)
+// elfv1-be-NEXT: stb [[REG1]], 2(3)
+// elfv1-be-NEXT: sth [[REG2]], 0(3)
+// elfv2-be: lhz [[REG1:.*]], 0(3)
+// elfv2-be-NEXT: lbz 3, 2(3)
+// elfv2-be-NEXT: rldimi 3, [[REG1]], 8, 0
+// elfv2-le: lbz [[REG1:.*]], 2(3)
+// elfv2-le-NEXT: lhz 3, 0(3)
+// elfv2-le-NEXT: rldimi 3, [[REG1]], 16, 0
+// CHECK-NEXT: blr
+#[no_mangle]
+extern "C" fn read_small(x: &ThreeU8s) -> ThreeU8s {
+    *x
+}
+
+// CHECK-LABEL: write_large
+// CHECK: std 3, 0(6)
+// be-NEXT: rldicl [[REG1:.*]], 5, 32, 32
+// CHECK-NEXT: std 4, 8(6)
+// be-NEXT: stw [[REG1]], 16(6)
+// elfv2-le-NEXT: stw 5, 16(6)
+// CHECK-NEXT: blr
+#[no_mangle]
+extern "C" fn write_large(x: FiveU32s, dest: &mut FiveU32s) {
+    *dest = x;
+}
+
+// CHECK-LABEL: write_medium
+// CHECK: std 3, 0(5)
+// be-NEXT: rldicl [[REG1:.*]], 4, 16, 48
+// be-NEXT: sth [[REG1]], 8(5)
+// elfv2-le-NEXT: sth 4, 8(5)
+// CHECK-NEXT: blr
+#[no_mangle]
+extern "C" fn write_medium(x: FiveU16s, dest: &mut FiveU16s) {
+    *dest = x;
+}
+
+// CHECK-LABEL: write_small
+// be: stb 3, 2(4)
+// be-NEXT: srwi [[REG1:.*]], 3, 8
+// be-NEXT: sth [[REG1]], 0(4)
+// The order these instructions are emitted in changed in LLVM 18.
+// elfv2-le-DAG: sth 3, 0(4)
+// elfv2-le-DAG: srwi [[REG1:.*]], 3, 16
+// elfv2-le-NEXT: stb [[REG1]], 2(4)
+// CHECK-NEXT: blr
+#[no_mangle]
+extern "C" fn write_small(x: ThreeU8s, dest: &mut ThreeU8s) {
+    *dest = x;
+}