Rollup merge of #127168 - DianQK:cast-size, r=workingjubilee

Use the aligned size for alloca at args/ret when the pass mode is cast

Fixes #75839. Fixes #121028.

The `load` and `store` instructions in LLVM access the aligned size. For example, `load { i64, i32 }` accesses 16 bytes on x86_64: https://alive2.llvm.org/ce/z/n8CHAp.

BTW, this example is expected to be optimized to immediate UB by Alive2: https://rust.godbolt.org/z/b7xK7hv1c and https://alive2.llvm.org/ce/z/vZDtZH.

r? compiler

compiler/rustc_codegen_llvm/src/abi.rs

@@ -226,7 +226,8 @@ impl<'ll, 'tcx> ArgAbiExt<'ll, 'tcx> for ArgAbi<'tcx, Ty<'tcx>> {
                 //   when passed by value, making it smaller.
                 // - On some ABIs, the Rust layout { u16, u16, u16 } may be padded up to 8 bytes
                 //   when passed by value, making it larger.
-                let copy_bytes = cmp::min(scratch_size.bytes(), self.layout.size.bytes());
+                let copy_bytes =
+                    cmp::min(cast.unaligned_size(bx).bytes(), self.layout.size.bytes());
                 // Allocate some scratch space...
                 let llscratch = bx.alloca(scratch_size, scratch_align);
                 bx.lifetime_start(llscratch, scratch_size);

compiler/rustc_codegen_ssa/src/mir/block.rs

@@ -1521,7 +1521,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                 //   when passed by value, making it smaller.
                 // - On some ABIs, the Rust layout { u16, u16, u16 } may be padded up to 8 bytes
                 //   when passed by value, making it larger.
-                let copy_bytes = cmp::min(scratch_size.bytes(), arg.layout.size.bytes());
+                let copy_bytes = cmp::min(cast.unaligned_size(bx).bytes(), arg.layout.size.bytes());
                 // Allocate some scratch space...
                 let llscratch = bx.alloca(scratch_size, scratch_align);
                 bx.lifetime_start(llscratch, scratch_size);

compiler/rustc_codegen_ssa/src/mir/mod.rs

@@ -230,10 +230,20 @@ pub fn codegen_mir<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
             let layout = start_bx.layout_of(fx.monomorphize(decl.ty));
             assert!(!layout.ty.has_erasable_regions());
 
-            if local == mir::RETURN_PLACE && fx.fn_abi.ret.is_indirect() {
+            if local == mir::RETURN_PLACE {
-                debug!("alloc: {:?} (return place) -> place", local);
+                match fx.fn_abi.ret.mode {
-                let llretptr = start_bx.get_param(0);
+                    PassMode::Indirect { .. } => {
-                return LocalRef::Place(PlaceRef::new_sized(llretptr, layout));
+                        debug!("alloc: {:?} (return place) -> place", local);
+                        let llretptr = start_bx.get_param(0);
+                        return LocalRef::Place(PlaceRef::new_sized(llretptr, layout));
+                    }
+                    PassMode::Cast { ref cast, .. } => {
+                        debug!("alloc: {:?} (return place) -> place", local);
+                        let size = cast.size(&start_bx);
+                        return LocalRef::Place(PlaceRef::alloca_size(&mut start_bx, size, layout));
+                    }
+                    _ => {}
+                };
             }
 
             if memory_locals.contains(local) {

compiler/rustc_codegen_ssa/src/mir/place.rs

@@ -108,9 +108,17 @@ impl<'a, 'tcx, V: CodegenObject> PlaceRef<'tcx, V> {
     pub fn alloca<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
         bx: &mut Bx,
         layout: TyAndLayout<'tcx>,
+    ) -> Self {
+        Self::alloca_size(bx, layout.size, layout)
+    }
+
+    pub fn alloca_size<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
+        bx: &mut Bx,
+        size: Size,
+        layout: TyAndLayout<'tcx>,
     ) -> Self {
         assert!(layout.is_sized(), "tried to statically allocate unsized place");
-        PlaceValue::alloca(bx, layout.size, layout.align.abi).with_type(layout)
+        PlaceValue::alloca(bx, size, layout.align.abi).with_type(layout)
     }
 
     /// Returns a place for an indirect reference to an unsized place.

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

@@ -339,7 +339,9 @@ impl CastTarget {
         }
     }
 
-    pub fn size<C: HasDataLayout>(&self, _cx: &C) -> Size {
+    /// When you only access the range containing valid data, you can use this unaligned size;
+    /// otherwise, use the safer `size` method.
+    pub fn unaligned_size<C: HasDataLayout>(&self, _cx: &C) -> Size {
         // Prefix arguments are passed in specific designated registers
         let prefix_size = self
             .prefix
@@ -353,6 +355,10 @@ impl CastTarget {
         prefix_size + rest_size
     }
 
+    pub fn size<C: HasDataLayout>(&self, cx: &C) -> Size {
+        self.unaligned_size(cx).align_to(self.align(cx))
+    }
+
     pub fn align<C: HasDataLayout>(&self, cx: &C) -> Align {
         self.prefix
             .iter()

tests/codegen/cast-target-abi.rs

@@ -1,6 +1,7 @@
 // ignore-tidy-linelength
-//@ revisions:aarch64 loongarch64 powerpc64 sparc64
+//@ revisions:aarch64 loongarch64 powerpc64 sparc64 x86_64
-//@ compile-flags: -O -C no-prepopulate-passes
+// FIXME: Add `-Cllvm-args=--lint-abort-on-error` after LLVM 19
+//@ compile-flags: -O -C no-prepopulate-passes -C passes=lint
 
 //@[aarch64] compile-flags: --target aarch64-unknown-linux-gnu
 //@[aarch64] needs-llvm-components: arm
@@ -10,6 +11,8 @@
 //@[powerpc64] needs-llvm-components: powerpc
 //@[sparc64] compile-flags: --target sparc64-unknown-linux-gnu
 //@[sparc64] needs-llvm-components: sparc
+//@[x86_64] compile-flags: --target x86_64-unknown-linux-gnu
+//@[x86_64] needs-llvm-components: x86
 
 // Tests that arguments with `PassMode::Cast` are handled correctly.
 
@@ -60,30 +63,261 @@ pub struct DoubleFloat {
     g: f32,
 }
 
-extern "C" {
+// On x86_64, this struct will be passed as `{ i64, i32 }`.
-    fn receives_twou16s(x: TwoU16s);
+// The load and store instructions will access 16 bytes, so we should allocate 16 bytes.
-    fn returns_twou16s() -> TwoU16s;
+#[repr(C)]
-
+pub struct Three32s {
-    fn receives_fiveu16s(x: FiveU16s);
+    a: u32,
-    fn returns_fiveu16s() -> FiveU16s;
+    b: u32,
-
+    c: u32,
-    fn receives_doubledouble(x: DoubleDouble);
-    fn returns_doubledouble() -> DoubleDouble;
-
-    // These functions cause an ICE in sparc64 ABI code (https://github.com/rust-lang/rust/issues/122620)
-    #[cfg(not(target_arch = "sparc64"))]
-    fn receives_doublefloat(x: DoubleFloat);
-    #[cfg(not(target_arch = "sparc64"))]
-    fn returns_doublefloat() -> DoubleFloat;
 }
 
-// CHECK-LABEL: @call_twou16s
+// CHECK-LABEL: @receives_twou16s
+// aarch64-SAME:     ([[ABI_TYPE:i64]] {{.*}}[[ABI_VALUE:%.+]])
+// loongarch64-SAME: ([[ABI_TYPE:i64]] {{.*}}[[ABI_VALUE:%.+]])
+// powerpc64-SAME:   ([[ABI_TYPE:i32]] {{.*}}[[ABI_VALUE:%.+]])
+// sparc64-SAME:     ([[ABI_TYPE:i64]] {{.*}}[[ABI_VALUE:%.+]])
+// x86_64-SAME:      ([[ABI_TYPE:i32]] {{.*}}[[ABI_VALUE:%.+]])
 #[no_mangle]
-pub unsafe fn call_twou16s() {
+#[inline(never)]
+pub extern "C" fn receives_twou16s(x: TwoU16s) {
     // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:8]]
     // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:8]]
     // powerpc64:   [[ABI_ALLOCA:%.+]] = alloca [4 x i8], align [[ABI_ALIGN:4]]
     // sparc64:     [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:8]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [4 x i8], align [[ABI_ALIGN:4]]
+
+    // CHECK: [[RUST_ALLOCA:%.+]] = alloca [4 x i8], align [[RUST_ALIGN:2]]
+
+    // CHECK: store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // CHECK: call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 4, i1 false)
+}
+
+// CHECK-LABEL: @returns_twou16s
+// powerpc64-SAME: sret([4 x i8]) align [[RUST_ALIGN:2]] {{.*}}[[RET_PTR:%.*]])
+#[no_mangle]
+#[inline(never)]
+pub extern "C" fn returns_twou16s() -> TwoU16s {
+    // powerpc returns this struct via sret pointer, it doesn't use the cast ABI.
+    // The other targets copy the cast ABI type to an alloca.
+
+    // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:2]]
+    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:2]]
+    // sparc64:     [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:2]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [4 x i8], align [[ABI_ALIGN:2]]
+
+    // aarch64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:i64]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // loongarch64: [[ABI_VALUE:%.+]] = load [[ABI_TYPE:i64]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // sparc64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:i64]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      [[ABI_VALUE:%.+]] = load [[ABI_TYPE:i32]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // aarch64:     ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // loongarch64: ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // sparc64:     ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // x86_64:      ret [[ABI_TYPE]] [[ABI_VALUE]]
+    TwoU16s { a: 0, b: 1 }
+}
+
+// CHECK-LABEL: @receives_fiveu16s
+// aarch64-SAME:     ([[ABI_TYPE:\[2 x i64\]]] {{.*}}[[ABI_VALUE:%.+]])
+// loongarch64-SAME: ([[ABI_TYPE:\[2 x i64\]]] {{.*}}[[ABI_VALUE:%.+]])
+// powerpc64-SAME:   ([[ABI_TYPE:\[2 x i64\]]] {{.*}}[[ABI_VALUE:%.+]])
+// sparc64-SAME:     ([[ABI_TYPE:\[2 x i64\]]] {{.*}}[[ABI_VALUE:%.+]])
+// x86_64-SAME:      ([[ABI_TYPE:{ i64, i16 }]] {{.*}}[[ABI_VALUE:%.+]])
+#[no_mangle]
+#[inline(never)]
+pub extern "C" fn receives_fiveu16s(x: FiveU16s) {
+    // CHECK: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+
+    // CHECK: [[RUST_ALLOCA:%.+]] = alloca [10 x i8], align [[RUST_ALIGN:2]]
+
+    // CHECK: store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // CHECK: call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 10, i1 false)
+}
+
+// CHECK-LABEL: @returns_fiveu16s
+// powerpc64-SAME: sret([10 x i8]) align [[RUST_ALIGN:2]] {{.*}}[[RET_PTR:%.*]])
+#[no_mangle]
+#[inline(never)]
+pub extern "C" fn returns_fiveu16s() -> FiveU16s {
+    // powerpc returns this struct via sret pointer, it doesn't use the cast ABI.
+    // The other targets copy the cast ABI type to an alloca.
+
+    // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:2]]
+    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:2]]
+    // sparc64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:2]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:2]]
+
+    // aarch64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // loongarch64: [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // sparc64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ i64, i16 }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // aarch64:     ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // loongarch64: ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // sparc64:     ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // x86_64:      ret [[ABI_TYPE]] [[ABI_VALUE]]
+    FiveU16s { a: 0, b: 1, c: 2, d: 3, e: 4 }
+}
+
+// CHECK-LABEL: @receives_doubledouble
+// aarch64-SAME:     ([[ABI_TYPE:\[2 x double\]]] {{.*}}[[ABI_VALUE:%.+]])
+// loongarch64-SAME: ([[ABI_TYPE:{ double, double }]] {{.*}}[[ABI_VALUE:%.+]])
+// powerpc64-SAME:   ([[ABI_TYPE:\[2 x i64\]]] {{.*}}[[ABI_VALUE:%.+]])
+// sparc64-SAME:     ([[ABI_TYPE:{ double, double }]] {{.*}}[[ABI_VALUE:%.+]])
+// x86_64-SAME:      ([[ABI_TYPE:{ double, double }]] {{.*}}[[ABI_VALUE:%.+]])
+#[no_mangle]
+#[inline(never)]
+pub extern "C" fn receives_doubledouble(x: DoubleDouble) {
+    // CHECK: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+
+    // CHECK: [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
+
+    // CHECK: store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // CHECK: call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 16, i1 false)
+}
+
+// CHECK-LABEL: @returns_doubledouble
+// powerpc64-SAME: sret([16 x i8]) align [[RUST_ALIGN:8]] {{.*}}[[RET_PTR:%.*]])
+#[no_mangle]
+#[inline(never)]
+pub extern "C" fn returns_doubledouble() -> DoubleDouble {
+    // powerpc returns this struct via sret pointer, it doesn't use the cast ABI.
+    // The other targets copy the cast ABI type to an alloca.
+
+    // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // sparc64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+
+    // aarch64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x double\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // loongarch64: [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ double, double }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // sparc64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ double, double }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ double, double }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // aarch64:     ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // loongarch64: ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // sparc64:     ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // x86_64:      ret [[ABI_TYPE]] [[ABI_VALUE]]
+    DoubleDouble { f: 0., g: 1. }
+}
+
+// CHECK-LABEL: @receives_three32s
+// aarch64-SAME:     ([[ABI_TYPE:\[2 x i64\]]] {{.*}}[[ABI_VALUE:%.+]])
+// loongarch64-SAME: ([[ABI_TYPE:\[2 x i64\]]] {{.*}}[[ABI_VALUE:%.+]])
+// powerpc64-SAME:   ([[ABI_TYPE:\[2 x i64\]]] {{.*}}[[ABI_VALUE:%.+]])
+// sparc64-SAME:     ([[ABI_TYPE:\[2 x i64\]]] {{.*}}[[ABI_VALUE:%.+]])
+// x86_64-SAME:      ([[ABI_TYPE:{ i64, i32 }]] {{.*}}[[ABI_VALUE:%.+]])
+#[no_mangle]
+#[inline(never)]
+pub extern "C" fn receives_three32s(x: Three32s) {
+    // CHECK: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+
+    // CHECK: [[RUST_ALLOCA:%.+]] = alloca [12 x i8], align [[RUST_ALIGN:4]]
+
+    // CHECK: store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // CHECK: call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 12, i1 false)
+}
+
+// CHECK-LABEL: @returns_three32s
+// powerpc64-SAME: sret([12 x i8]) align [[RUST_ALIGN:4]] {{.*}}[[RET_PTR:%.*]])
+#[no_mangle]
+#[inline(never)]
+pub extern "C" fn returns_three32s() -> Three32s {
+    // powerpc returns this struct via sret pointer, it doesn't use the cast ABI.
+    // The other targets copy the cast ABI type to an alloca.
+
+    // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:4]]
+    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:4]]
+    // sparc64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:4]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:4]]
+
+    // aarch64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // loongarch64: [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // sparc64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ i64, i32 }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // aarch64:     ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // loongarch64: ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // sparc64:     ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // x86_64:      ret [[ABI_TYPE]] [[ABI_VALUE]]
+    Three32s { a: 0, b: 0, c: 0 }
+}
+
+// These functions cause an ICE in sparc64 ABI code (https://github.com/rust-lang/rust/issues/122620)
+#[cfg(not(target_arch = "sparc64"))]
+// aarch64-LABEL:     @receives_doublefloat
+// loongarch64-LABEL: @receives_doublefloat
+// powerpc64-LABEL:   @receives_doublefloat
+// x86_64-LABEL:      @receives_doublefloat
+
+// aarch64-SAME:     ([[ABI_TYPE:\[2 x i64\]]] {{.*}}[[ABI_VALUE:%.+]])
+// loongarch64-SAME: ([[ABI_TYPE:{ double, float }]] {{.*}}[[ABI_VALUE:%.+]])
+// powerpc64-SAME:   ([[ABI_TYPE:\[2 x i64\]]] {{.*}}[[ABI_VALUE:%.+]])
+// x86_64-SAME:      ([[ABI_TYPE:{ double, double }]] {{.*}}[[ABI_VALUE:%.+]])
+#[no_mangle]
+#[inline(never)]
+pub extern "C" fn receives_doublefloat(x: DoubleFloat) {
+    // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // powerpc64:   [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+
+    // aarch64:     [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
+    // loongarch64: [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
+    // powerpc64:   [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
+    // x86_64:      [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
+
+    // aarch64:     store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // loongarch64: store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // powerpc64:   store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // aarch64:     call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 16, i1 false)
+    // loongarch64: call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 12, i1 false)
+    // powerpc64:   call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 16, i1 false)
+    // x86_64:      call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 16, i1 false)
+}
+
+#[cfg(not(target_arch = "sparc64"))]
+// aarch64-LABEL:     @returns_doublefloat
+// loongarch64-LABEL: @returns_doublefloat
+// powerpc64-LABEL:   @returns_doublefloat
+// x86_64-LABEL:      @returns_doublefloat
+
+// powerpc64-SAME: sret([16 x i8]) align [[RUST_ALIGN:8]] {{.*}}[[RET_PTR:%.*]])
+#[no_mangle]
+#[inline(never)]
+pub extern "C" fn returns_doublefloat() -> DoubleFloat {
+    // powerpc returns this struct via sret pointer, it doesn't use the cast ABI.
+    // The other targets copy the cast ABI type to an alloca.
+
+    // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+
+    // aarch64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // loongarch64: [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ double, float }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ double, double }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // aarch64:     ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // loongarch64: ret [[ABI_TYPE]] [[ABI_VALUE]]
+    // x86_64:      ret [[ABI_TYPE]] [[ABI_VALUE]]
+    DoubleFloat { f: 0., g: 0. }
+}
+
+// CHECK-LABEL: @call_twou16s
+#[no_mangle]
+pub fn call_twou16s() {
+    // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:8]]
+    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:8]]
+    // powerpc64:   [[ABI_ALLOCA:%.+]] = alloca [4 x i8], align [[ABI_ALIGN:4]]
+    // sparc64:     [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:8]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [4 x i8], align [[ABI_ALIGN:4]]
 
     // CHECK: [[RUST_ALLOCA:%.+]] = alloca [4 x i8], align [[RUST_ALIGN:2]]
 
@@ -93,6 +327,7 @@ pub unsafe fn call_twou16s() {
     // loongarch64: [[ABI_VALUE:%.+]] = load [[ABI_TYPE:i64]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // powerpc64:   [[ABI_VALUE:%.+]] = load [[ABI_TYPE:i32]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // sparc64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:i64]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      [[ABI_VALUE:%.+]] = load [[ABI_TYPE:i32]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
 
     // CHECK: call void @receives_twou16s([[ABI_TYPE]] [[ABI_VALUE]])
     let x = TwoU16s { a: 1, b: 2 };
@@ -101,7 +336,7 @@ pub unsafe fn call_twou16s() {
 
 // CHECK-LABEL: @return_twou16s
 #[no_mangle]
-pub unsafe fn return_twou16s() -> TwoU16s {
+pub fn return_twou16s() -> TwoU16s {
     // powerpc returns this struct via sret pointer, it doesn't use the cast ABI.
 
     // powerpc64: [[RETVAL:%.+]] = alloca [4 x i8], align 2
@@ -112,34 +347,45 @@ pub unsafe fn return_twou16s() -> TwoU16s {
     // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:8]]
     // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:8]]
     // sparc64:     [[ABI_ALLOCA:%.+]] = alloca [8 x i8], align [[ABI_ALIGN:8]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [4 x i8], align [[ABI_ALIGN:4]]
 
     // aarch64:     [[RUST_ALLOCA:%.+]] = alloca [4 x i8], align [[RUST_ALIGN:2]]
     // loongarch64: [[RUST_ALLOCA:%.+]] = alloca [4 x i8], align [[RUST_ALIGN:2]]
     // sparc64:     [[RUST_ALLOCA:%.+]] = alloca [4 x i8], align [[RUST_ALIGN:2]]
+    // x86_64:      [[RUST_ALLOCA:%.+]] = alloca [4 x i8], align [[RUST_ALIGN:2]]
 
     // aarch64:     [[ABI_VALUE:%.+]] = call [[ABI_TYPE:i64]] @returns_twou16s()
     // loongarch64: [[ABI_VALUE:%.+]] = call [[ABI_TYPE:i64]] @returns_twou16s()
     // sparc64:     [[ABI_VALUE:%.+]] = call [[ABI_TYPE:i64]] @returns_twou16s()
+    // x86_64:      [[ABI_VALUE:%.+]] = call [[ABI_TYPE:i32]] @returns_twou16s()
 
     // aarch64:     store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // loongarch64: store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // sparc64:     store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
 
     // aarch64:     call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 4, i1 false)
     // loongarch64: call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 4, i1 false)
     // sparc64:     call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 4, i1 false)
+    // x86_64:      call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 4, i1 false)
     returns_twou16s()
 }
 
 // CHECK-LABEL: @call_fiveu16s
 #[no_mangle]
-pub unsafe fn call_fiveu16s() {
+pub fn call_fiveu16s() {
     // CHECK: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
 
     // CHECK: [[RUST_ALLOCA:%.+]] = alloca [10 x i8], align 2
 
     // CHECK: call void @llvm.memcpy.{{.+}}(ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], i64 10, i1 false)
-    // CHECK: [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // aarch64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // loongarch64: [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // powerpc64:   [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // sparc64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ i64, i16 }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
     // CHECK: call void @receives_fiveu16s([[ABI_TYPE]] [[ABI_VALUE]])
     let x = FiveU16s { a: 1, b: 2, c: 3, d: 4, e: 5 };
     receives_fiveu16s(x);
@@ -148,7 +394,7 @@ pub unsafe fn call_fiveu16s() {
 // CHECK-LABEL: @return_fiveu16s
 // CHECK-SAME: (ptr {{.+}} sret([10 x i8]) align [[RUST_ALIGN:2]] dereferenceable(10) [[RET_PTR:%.+]])
 #[no_mangle]
-pub unsafe fn return_fiveu16s() -> FiveU16s {
+pub fn return_fiveu16s() -> FiveU16s {
     // powerpc returns this struct via sret pointer, it doesn't use the cast ABI.
 
     // powerpc64: call void @returns_fiveu16s(ptr {{.+}} [[RET_PTR]])
@@ -158,24 +404,28 @@ pub unsafe fn return_fiveu16s() -> FiveU16s {
     // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
     // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
     // sparc64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
 
     // aarch64:     [[ABI_VALUE:%.+]] = call [[ABI_TYPE:\[2 x i64\]]] @returns_fiveu16s()
     // loongarch64: [[ABI_VALUE:%.+]] = call [[ABI_TYPE:\[2 x i64\]]] @returns_fiveu16s()
     // sparc64:     [[ABI_VALUE:%.+]] = call [[ABI_TYPE:\[2 x i64\]]] @returns_fiveu16s()
+    // x86_64:      [[ABI_VALUE:%.+]] = call [[ABI_TYPE:{ i64, i16 }]] @returns_fiveu16s()
 
     // aarch64:     store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // loongarch64: store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // sparc64:     store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
 
     // aarch64:     call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RET_PTR]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 10, i1 false)
     // loongarch64: call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RET_PTR]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 10, i1 false)
     // sparc64:     call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RET_PTR]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 10, i1 false)
+    // x86_64:      call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RET_PTR]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 10, i1 false)
     returns_fiveu16s()
 }
 
 // CHECK-LABEL: @call_doubledouble
 #[no_mangle]
-pub unsafe fn call_doubledouble() {
+pub fn call_doubledouble() {
     // CHECK: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
 
     // CHECK: [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
@@ -186,6 +436,7 @@ pub unsafe fn call_doubledouble() {
     // loongarch64: [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ double, double }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // powerpc64:   [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // sparc64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ double, double }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ double, double }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
 
     // CHECK: call void @receives_doubledouble([[ABI_TYPE]] [[ABI_VALUE]])
     let x = DoubleDouble { f: 1., g: 2. };
@@ -194,7 +445,7 @@ pub unsafe fn call_doubledouble() {
 
 // CHECK-LABEL: @return_doubledouble
 #[no_mangle]
-pub unsafe fn return_doubledouble() -> DoubleDouble {
+pub fn return_doubledouble() -> DoubleDouble {
     // powerpc returns this struct via sret pointer, it doesn't use the cast ABI.
 
     // powerpc64: [[RETVAL:%.+]] = alloca [16 x i8], align 8
@@ -205,22 +456,27 @@ pub unsafe fn return_doubledouble() -> DoubleDouble {
     // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
     // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
     // sparc64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
 
     // aarch64:     [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
     // loongarch64: [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
     // sparc64:     [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
+    // x86_64:      [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
 
     // aarch64:     [[ABI_VALUE:%.+]] = call [[ABI_TYPE:\[2 x double\]]] @returns_doubledouble()
     // loongarch64: [[ABI_VALUE:%.+]] = call [[ABI_TYPE:{ double, double }]] @returns_doubledouble()
     // sparc64:     [[ABI_VALUE:%.+]] = call [[ABI_TYPE:{ double, double }]] @returns_doubledouble()
+    // x86_64:      [[ABI_VALUE:%.+]] = call [[ABI_TYPE:{ double, double }]] @returns_doubledouble()
 
     // aarch64:     store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // loongarch64: store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // sparc64:     store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
 
     // aarch64:     call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 16, i1 false)
     // loongarch64: call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 16, i1 false)
     // sparc64:     call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 16, i1 false)
+    // x86_64:      call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 16, i1 false)
     returns_doubledouble()
 }
 
@@ -229,27 +485,33 @@ pub unsafe fn return_doubledouble() -> DoubleDouble {
 // aarch64-LABEL:     @call_doublefloat
 // loongarch64-LABEL: @call_doublefloat
 // powerpc64-LABEL:   @call_doublefloat
+// x86_64-LABEL:      @call_doublefloat
 #[no_mangle]
-pub unsafe fn call_doublefloat() {
+pub fn call_doublefloat() {
     // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
-    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [12 x i8], align [[ABI_ALIGN:8]]
+    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
     // powerpc64:   [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
 
     // aarch64:     [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
     // loongarch64: [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
     // powerpc64:   [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
+    // x86_64:      [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
 
     // aarch64:     call void @llvm.memcpy.{{.+}}(ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], i64 16, i1 false)
     // loongarch64: call void @llvm.memcpy.{{.+}}(ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], i64 12, i1 false)
     // powerpc64:   call void @llvm.memcpy.{{.+}}(ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], i64 16, i1 false)
+    // x86_64:      call void @llvm.memcpy.{{.+}}(ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], i64 16, i1 false)
 
     // aarch64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // loongarch64: [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ double, float }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // powerpc64:   [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ double, double }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
 
     // aarch64:     call void @receives_doublefloat([[ABI_TYPE]] {{(inreg )?}}[[ABI_VALUE]])
     // loongarch64: call void @receives_doublefloat([[ABI_TYPE]] {{(inreg )?}}[[ABI_VALUE]])
     // powerpc64:   call void @receives_doublefloat([[ABI_TYPE]] {{(inreg )?}}[[ABI_VALUE]])
+    // x86_64:      call void @receives_doublefloat([[ABI_TYPE]] {{(inreg )?}}[[ABI_VALUE]])
     let x = DoubleFloat { f: 1., g: 2. };
     receives_doublefloat(x);
 }
@@ -259,8 +521,9 @@ pub unsafe fn call_doublefloat() {
 // aarch64-LABEL:     @return_doublefloat
 // loongarch64-LABEL: @return_doublefloat
 // powerpc64-LABEL:   @return_doublefloat
+// x86_64-LABEL:      @return_doublefloat
 #[no_mangle]
-pub unsafe fn return_doublefloat() -> DoubleFloat {
+pub fn return_doublefloat() -> DoubleFloat {
     // powerpc returns this struct via sret pointer, it doesn't use the cast ABI.
 
     // powerpc64: [[RETVAL:%.+]] = alloca [16 x i8], align 8
@@ -269,18 +532,72 @@ pub unsafe fn return_doublefloat() -> DoubleFloat {
     // The other targets copy the cast ABI type to an alloca.
 
     // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
-    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [12 x i8], align [[ABI_ALIGN:8]]
+    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
 
     // aarch64:     [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
     // loongarch64: [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
+    // x86_64:      [[RUST_ALLOCA:%.+]] = alloca [16 x i8], align [[RUST_ALIGN:8]]
 
     // aarch64:     [[ABI_VALUE:%.+]] = call [[ABI_TYPE:\[2 x i64\]]] @returns_doublefloat()
     // loongarch64: [[ABI_VALUE:%.+]] = call [[ABI_TYPE:{ double, float }]] @returns_doublefloat()
+    // x86_64:      [[ABI_VALUE:%.+]] = call [[ABI_TYPE:{ double, double }]] @returns_doublefloat()
 
     // aarch64:     store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
     // loongarch64: store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
 
     // aarch64:     call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 16, i1 false)
     // loongarch64: call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 12, i1 false)
+    // x86_64:      call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 16, i1 false)
     returns_doublefloat()
 }
+
+// CHECK-LABEL: @call_three32s
+#[no_mangle]
+pub fn call_three32s() {
+    // CHECK: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // CHECK: [[RUST_ALLOCA:%.+]] = alloca [12 x i8], align [[RUST_ALIGN:4]]
+    // CHECK: call void @llvm.memcpy.{{.+}}(ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], ptr align [[RUST_ALIGN]] [[RUST_ALLOCA]], i64 12, i1 false)
+
+    // aarch64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // loongarch64: [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // powerpc64:   [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // sparc64:     [[ABI_VALUE:%.+]] = load [[ABI_TYPE:\[2 x i64\]]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      [[ABI_VALUE:%.+]] = load [[ABI_TYPE:{ i64, i32 }]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // CHECK: call void @receives_three32s([[ABI_TYPE]] [[ABI_VALUE]])
+    let x = Three32s { a: 1, b: 2, c: 3 };
+    receives_three32s(x);
+}
+
+// Regression test for #75839
+// CHECK-LABEL: @return_three32s(
+// CHECK-SAME: sret([12 x i8]) align [[RUST_ALIGN:4]] {{.*}}[[RUST_RETVAL:%.*]])
+#[no_mangle]
+pub fn return_three32s() -> Three32s {
+    // powerpc returns this struct via sret pointer, it doesn't use the cast ABI.
+
+    // powerpc64: call void @returns_three32s(ptr {{.+}} [[RUST_RETVAL]])
+
+    // aarch64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // loongarch64: [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // sparc64:     [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+    // x86_64:      [[ABI_ALLOCA:%.+]] = alloca [16 x i8], align [[ABI_ALIGN:8]]
+
+    // aarch64:     [[ABI_VALUE:%.+]] = call [[ABI_TYPE:\[2 x i64\]]] @returns_three32s()
+    // loongarch64: [[ABI_VALUE:%.+]] = call [[ABI_TYPE:\[2 x i64\]]] @returns_three32s()
+    // sparc64:     [[ABI_VALUE:%.+]] = call [[ABI_TYPE:\[2 x i64\]]] @returns_three32s()
+    // x86_64:      [[ABI_VALUE:%.+]] = call [[ABI_TYPE:{ i64, i32 }]] @returns_three32s()
+
+    // aarch64:     store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // loongarch64: store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // sparc64:     store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+    // x86_64:      store [[ABI_TYPE]] [[ABI_VALUE]], ptr [[ABI_ALLOCA]], align [[ABI_ALIGN]]
+
+    // aarch64:     call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_RETVAL]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 12, i1 false)
+    // loongarch64: call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_RETVAL]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 12, i1 false)
+    // sparc64:     call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_RETVAL]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 12, i1 false)
+    // x86_64:      call void @llvm.memcpy.{{.+}}(ptr align [[RUST_ALIGN]] [[RUST_RETVAL]], ptr align [[ABI_ALIGN]] [[ABI_ALLOCA]], i64 12, i1 false)
+    returns_three32s()
+}

tests/codegen/cffi/ffi-out-of-bounds-loads.rs

@@ -1,5 +1,5 @@
 //@ revisions: linux apple
-//@ compile-flags: -C opt-level=0 -C no-prepopulate-passes
+//@ compile-flags: -C opt-level=0 -C no-prepopulate-passes -C passes=lint
 
 //@[linux] compile-flags: --target x86_64-unknown-linux-gnu
 //@[linux] needs-llvm-components: x86
@@ -36,7 +36,7 @@ extern "C" {
 pub fn test() {
     let s = S { f1: 1, f2: 2, f3: 3 };
     unsafe {
-        // CHECK: [[ALLOCA:%.+]] = alloca [12 x i8], align 8
+        // CHECK: [[ALLOCA:%.+]] = alloca [16 x i8], align 8
         // CHECK: [[LOAD:%.+]] = load { i64, i32 }, ptr [[ALLOCA]], align 8
         // CHECK: call void @foo({ i64, i32 } [[LOAD]])
         foo(s);