Auto merge of #118991 - nikic:scalar-pair, r=<try>

Separate immediate and in-memory ScalarPair representation

Currently, we assume that ScalarPair is always represented using a two-element struct, both as an immediate value and when stored in memory.

This currently works fairly well, but runs into problems with #116672, where a ScalarPair involving an i128 type can no longer be represented as a two-element struct in memory. For example, the tuple `(i32, i128)` needs to be represented in-memory as `{ i32, [3 x i32], i128 }` to satisfy alignment requirements. Using `{ i32, i128 }` instead will result in the second element being stored at the wrong offset (prior to LLVM 18).

Resolve this issue by no longer requiring that the immediate and in-memory type for ScalarPair are the same. The in-memory type will now look the same as for normal struct types (and will include padding filler and similar), while the immediate type stays a simple two-element struct type. This also means that booleans in immediate ScalarPair are now represented as i1 rather than i8, just like we do everywhere else.

The core change here is to llvm_type (which now treats ScalarPair as a normal struct) and immediate_llvm_type (which returns the two-element struct that llvm_type used to produce). The rest is fixing things up to no longer assume these are the same. In particular, this switches places that try to get pointers to the ScalarPair elements to use byte-geps instead of struct-geps.

compiler/rustc_codegen_llvm/src/abi.rs

@@ -6,7 +6,7 @@ use crate::type_::Type;
 use crate::type_of::LayoutLlvmExt;
 use crate::value::Value;
 
-use rustc_codegen_ssa::mir::operand::OperandValue;
+use rustc_codegen_ssa::mir::operand::{OperandRef, OperandValue};
 use rustc_codegen_ssa::mir::place::PlaceRef;
 use rustc_codegen_ssa::traits::*;
 use rustc_codegen_ssa::MemFlags;
@@ -253,7 +253,7 @@ impl<'ll, 'tcx> ArgAbiExt<'ll, 'tcx> for ArgAbi<'tcx, Ty<'tcx>> {
                 bx.lifetime_end(llscratch, scratch_size);
             }
         } else {
-            OperandValue::Immediate(val).store(bx, dst);
+            OperandRef::from_immediate_or_packed_pair(bx, val, self.layout).val.store(bx, dst);
         }
     }
 

compiler/rustc_codegen_llvm/src/builder.rs

@@ -558,10 +558,17 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
             OperandValue::Immediate(self.to_immediate(llval, place.layout))
         } else if let abi::Abi::ScalarPair(a, b) = place.layout.abi {
             let b_offset = a.size(self).align_to(b.align(self).abi);
-            let pair_ty = place.layout.llvm_type(self);
 
             let mut load = |i, scalar: abi::Scalar, layout, align, offset| {
-                let llptr = self.struct_gep(pair_ty, place.llval, i as u64);
+                let llptr = if i == 0 {
+                    place.llval
+                } else {
+                    self.inbounds_gep(
+                        self.type_i8(),
+                        place.llval,
+                        &[self.const_usize(b_offset.bytes())],
+                    )
+                };
                 let llty = place.layout.scalar_pair_element_llvm_type(self, i, false);
                 let load = self.load(llty, llptr, align);
                 scalar_load_metadata(self, load, scalar, layout, offset);

compiler/rustc_codegen_llvm/src/intrinsic.rs

@@ -179,7 +179,10 @@ impl<'ll, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                 unsafe {
                     llvm::LLVMSetAlignment(load, align);
                 }
-                self.to_immediate(load, self.layout_of(tp_ty))
+                if !result.layout.is_zst() {
+                    self.store(load, result.llval, result.align);
+                }
+                return;
             }
             sym::volatile_store => {
                 let dst = args[0].deref(self.cx());

compiler/rustc_codegen_llvm/src/type_of.rs

@@ -26,16 +26,8 @@ fn uncached_llvm_type<'a, 'tcx>(
             let element = layout.scalar_llvm_type_at(cx, element);
             return cx.type_vector(element, count);
         }
-        Abi::ScalarPair(..) => {
-            return cx.type_struct(
-                &[
-                    layout.scalar_pair_element_llvm_type(cx, 0, false),
-                    layout.scalar_pair_element_llvm_type(cx, 1, false),
-                ],
-                false,
-            );
-        }
-        Abi::Uninhabited | Abi::Aggregate { .. } => {}
+        // Treat ScalarPair like a normal aggregate for the purposes of in-memory representation.
+        Abi::Uninhabited | Abi::Aggregate { .. } | Abi::ScalarPair(..) => {}
     }
 
     let name = match layout.ty.kind() {
@@ -275,11 +267,25 @@ impl<'tcx> LayoutLlvmExt<'tcx> for TyAndLayout<'tcx> {
     }
 
     fn immediate_llvm_type<'a>(&self, cx: &CodegenCx<'a, 'tcx>) -> &'a Type {
-        if let Abi::Scalar(scalar) = self.abi {
-            if scalar.is_bool() {
-                return cx.type_i1();
+        match self.abi {
+            Abi::Scalar(scalar) => {
+                if scalar.is_bool() {
+                    return cx.type_i1();
+                }
             }
-        }
+            Abi::ScalarPair(..) => {
+                // An immediate pair always contains just the two elements, without any padding
+                // filler, as it should never be stored to memory.
+                return cx.type_struct(
+                    &[
+                        self.scalar_pair_element_llvm_type(cx, 0, true),
+                        self.scalar_pair_element_llvm_type(cx, 1, true),
+                    ],
+                    false,
+                );
+            }
+            _ => {}
+        };
         self.llvm_type(cx)
     }
 

compiler/rustc_codegen_ssa/src/mir/operand.rs

@@ -231,14 +231,12 @@ impl<'a, 'tcx, V: CodegenObject> OperandRef<'tcx, V> {
         bx: &mut Bx,
     ) -> V {
         if let OperandValue::Pair(a, b) = self.val {
-            let llty = bx.cx().backend_type(self.layout);
+            let llty = bx.cx().immediate_backend_type(self.layout);
             debug!("Operand::immediate_or_packed_pair: packing {:?} into {:?}", self, llty);
             // Reconstruct the immediate aggregate.
             let mut llpair = bx.cx().const_poison(llty);
-            let imm_a = bx.from_immediate(a);
-            let imm_b = bx.from_immediate(b);
-            llpair = bx.insert_value(llpair, imm_a, 0);
-            llpair = bx.insert_value(llpair, imm_b, 1);
+            llpair = bx.insert_value(llpair, a, 0);
+            llpair = bx.insert_value(llpair, b, 1);
             llpair
         } else {
             self.immediate()
@@ -251,14 +249,12 @@ impl<'a, 'tcx, V: CodegenObject> OperandRef<'tcx, V> {
         llval: V,
         layout: TyAndLayout<'tcx>,
     ) -> Self {
-        let val = if let Abi::ScalarPair(a, b) = layout.abi {
+        let val = if let Abi::ScalarPair(..) = layout.abi {
             debug!("Operand::from_immediate_or_packed_pair: unpacking {:?} @ {:?}", llval, layout);
 
             // Deconstruct the immediate aggregate.
             let a_llval = bx.extract_value(llval, 0);
-            let a_llval = bx.to_immediate_scalar(a_llval, a);
             let b_llval = bx.extract_value(llval, 1);
-            let b_llval = bx.to_immediate_scalar(b_llval, b);
             OperandValue::Pair(a_llval, b_llval)
         } else {
             OperandValue::Immediate(llval)
@@ -435,15 +431,14 @@ impl<'a, 'tcx, V: CodegenObject> OperandValue<V> {
                 let Abi::ScalarPair(a_scalar, b_scalar) = dest.layout.abi else {
                     bug!("store_with_flags: invalid ScalarPair layout: {:#?}", dest.layout);
                 };
-                let ty = bx.backend_type(dest.layout);
                 let b_offset = a_scalar.size(bx).align_to(b_scalar.align(bx).abi);
 
-                let llptr = bx.struct_gep(ty, dest.llval, 0);
                 let val = bx.from_immediate(a);
                 let align = dest.align;
-                bx.store_with_flags(val, llptr, align, flags);
+                bx.store_with_flags(val, dest.llval, align, flags);
 
-                let llptr = bx.struct_gep(ty, dest.llval, 1);
+                let llptr =
+                    bx.inbounds_gep(bx.type_i8(), dest.llval, &[bx.const_usize(b_offset.bytes())]);
                 let val = bx.from_immediate(b);
                 let align = dest.align.restrict_for_offset(b_offset);
                 bx.store_with_flags(val, llptr, align, flags);

compiler/rustc_codegen_ssa/src/mir/place.rs

@@ -112,8 +112,7 @@ impl<'a, 'tcx, V: CodegenObject> PlaceRef<'tcx, V> {
                     if offset == a.size(bx.cx()).align_to(b.align(bx.cx()).abi) =>
                 {
                     // Offset matches second field.
-                    let ty = bx.backend_type(self.layout);
-                    bx.struct_gep(ty, self.llval, 1)
+                    bx.inbounds_gep(bx.type_i8(), self.llval, &[bx.const_usize(offset.bytes())])
                 }
                 Abi::Scalar(_) | Abi::ScalarPair(..) | Abi::Vector { .. } if field.is_zst() => {
                     // ZST fields (even some that require alignment) are not included in Scalar,

tests/codegen/align-byval.rs

@@ -106,21 +106,21 @@ pub struct ForceAlign16 {
 pub unsafe fn call_na1(x: NaturalAlign1) {
     // CHECK: start:
 
-    // m68k: [[ALLOCA:%[a-z0-9+]]] = alloca { i8, i8 }, align 1
-    // m68k: call void @natural_align_1({{.*}}byval({ i8, i8 }) align 1{{.*}} [[ALLOCA]])
+    // m68k: [[ALLOCA:%[a-z0-9+]]] = alloca %NaturalAlign1, align 1
+    // m68k: call void @natural_align_1({{.*}}byval(%NaturalAlign1) align 1{{.*}} [[ALLOCA]])
 
-    // wasm: [[ALLOCA:%[a-z0-9+]]] = alloca { i8, i8 }, align 1
-    // wasm: call void @natural_align_1({{.*}}byval({ i8, i8 }) align 1{{.*}} [[ALLOCA]])
+    // wasm: [[ALLOCA:%[a-z0-9+]]] = alloca %NaturalAlign1, align 1
+    // wasm: call void @natural_align_1({{.*}}byval(%NaturalAlign1) align 1{{.*}} [[ALLOCA]])
 
     // x86_64-linux: call void @natural_align_1(i16
 
     // x86_64-windows: call void @natural_align_1(i16
 
-    // i686-linux: [[ALLOCA:%[a-z0-9+]]] = alloca { i8, i8 }, align 4
-    // i686-linux: call void @natural_align_1({{.*}}byval({ i8, i8 }) align 4{{.*}} [[ALLOCA]])
+    // i686-linux: [[ALLOCA:%[a-z0-9+]]] = alloca %NaturalAlign1, align 4
+    // i686-linux: call void @natural_align_1({{.*}}byval(%NaturalAlign1) align 4{{.*}} [[ALLOCA]])
 
-    // i686-windows: [[ALLOCA:%[a-z0-9+]]] = alloca { i8, i8 }, align 4
-    // i686-windows: call void @natural_align_1({{.*}}byval({ i8, i8 }) align 4{{.*}} [[ALLOCA]])
+    // i686-windows: [[ALLOCA:%[a-z0-9+]]] = alloca %NaturalAlign1, align 4
+    // i686-windows: call void @natural_align_1({{.*}}byval(%NaturalAlign1) align 4{{.*}} [[ALLOCA]])
     natural_align_1(x);
 }
 
@@ -199,17 +199,17 @@ pub unsafe fn call_fa16(x: ForceAlign16) {
 }
 
 extern "C" {
-    // m68k: declare void @natural_align_1({{.*}}byval({ i8, i8 }) align 1{{.*}})
+    // m68k: declare void @natural_align_1({{.*}}byval(%NaturalAlign1) align 1{{.*}})
 
-    // wasm: declare void @natural_align_1({{.*}}byval({ i8, i8 }) align 1{{.*}})
+    // wasm: declare void @natural_align_1({{.*}}byval(%NaturalAlign1) align 1{{.*}})
 
     // x86_64-linux: declare void @natural_align_1(i16)
 
     // x86_64-windows: declare void @natural_align_1(i16)
 
-    // i686-linux: declare void @natural_align_1({{.*}}byval({ i8, i8 }) align 4{{.*}})
+    // i686-linux: declare void @natural_align_1({{.*}}byval(%NaturalAlign1) align 4{{.*}})
 
-    // i686-windows: declare void @natural_align_1({{.*}}byval({ i8, i8 }) align 4{{.*}})
+    // i686-windows: declare void @natural_align_1({{.*}}byval(%NaturalAlign1) align 4{{.*}})
     fn natural_align_1(x: NaturalAlign1);
 
     // m68k: declare void @natural_align_2({{.*}}byval(%NaturalAlign2) align 2{{.*}})

tests/codegen/align-struct.rs

@@ -57,7 +57,7 @@ pub fn nested64(a: Align64, b: i32, c: i32, d: i8) -> Nested64 {
 // CHECK-LABEL: @enum4
 #[no_mangle]
 pub fn enum4(a: i32) -> Enum4 {
-// CHECK: %e4 = alloca { i32, i32 }, align 4
+// CHECK: %e4 = alloca %Enum4, align 4
     let e4 = Enum4::A(a);
     e4
 }

tests/codegen/function-arguments-noopt.rs

@@ -55,7 +55,7 @@ pub fn struct_call(x: S, f: fn(S) -> S) -> S {
   f(x)
 }
 
-// CHECK: { i8, i8 } @enum_(i1 zeroext %x.0, i8 %x.1)
+// CHECK: { i1, i8 } @enum_(i1 zeroext %x.0, i8 %x.1)
 #[no_mangle]
 pub fn enum_(x: Option<u8>) -> Option<u8> {
   x
@@ -64,6 +64,6 @@ pub fn enum_(x: Option<u8>) -> Option<u8> {
 // CHECK-LABEL: @enum_call
 #[no_mangle]
 pub fn enum_call(x: Option<u8>, f: fn(Option<u8>) -> Option<u8>) -> Option<u8> {
-  // CHECK: call { i8, i8 } %f(i1 zeroext %x.0, i8 %x.1)
+  // CHECK: call { i1, i8 } %f(i1 zeroext %x.0, i8 %x.1)
   f(x)
 }

tests/codegen/function-arguments.rs

@@ -275,7 +275,7 @@ pub fn enum_id_1(x: Option<Result<u16, u16>>) -> Option<Result<u16, u16>> {
   x
 }
 
-// CHECK: { i8, i8 } @enum_id_2(i1 noundef zeroext %x.0, i8 %x.1)
+// CHECK: { i1, i8 } @enum_id_2(i1 noundef zeroext %x.0, i8 %x.1)
 #[no_mangle]
 pub fn enum_id_2(x: Option<u8>) -> Option<u8> {
   x

tests/codegen/intrinsics/transmute.rs

@@ -190,7 +190,7 @@ pub unsafe fn check_byte_from_bool(x: bool) -> u8 {
 // CHECK-LABEL: @check_to_pair(
 #[no_mangle]
 pub unsafe fn check_to_pair(x: u64) -> Option<i32> {
-    // CHECK: %_0 = alloca { i32, i32 }, align 4
+    // CHECK: %_0 = alloca %"core::option::Option<i32>", align 4
     // CHECK: store i64 %x, ptr %_0, align 4
     transmute(x)
 }
@@ -203,10 +203,10 @@ pub unsafe fn check_from_pair(x: Option<i32>) -> u64 {
     const { assert!(std::mem::align_of::<Option<i32>>() == 4) };
 
     // CHECK: %_0 = alloca i64, align 8
-    // CHECK: store i32 %x.0, ptr %0, align 8
-    // CHECK: store i32 %x.1, ptr %1, align 4
-    // CHECK: %2 = load i64, ptr %_0, align 8
-    // CHECK: ret i64 %2
+    // CHECK: store i32 %x.0, ptr %_0, align 8
+    // CHECK: store i32 %x.1, ptr %0, align 4
+    // CHECK: %[[R:.+]] = load i64, ptr %_0, align 8
+    // CHECK: ret i64 %[[R]]
     transmute(x)
 }
 

tests/codegen/personality_lifetimes.rs

@@ -24,7 +24,7 @@ pub fn test() {
     let _s = S;
     // Check that the personality slot alloca gets a lifetime start in each cleanup block, not just
     // in the first one.
-    // CHECK: [[SLOT:%[0-9]+]] = alloca { ptr, i32 }
+    // CHECK: [[SLOT:%[0-9]+]] = alloca { ptr, i32, [1 x i32] }
     // CHECK-LABEL: cleanup:
     // CHECK: call void @llvm.lifetime.start.{{.*}}({{.*}})
     // CHECK-LABEL: cleanup1:

tests/codegen/refs.rs

@@ -13,9 +13,8 @@ pub fn helper(_: usize) {
 pub fn ref_dst(s: &[u8]) {
     // We used to generate an extra alloca and memcpy to ref the dst, so check that we copy
     // directly to the alloca for "x"
-// CHECK: [[X0:%[0-9]+]] = getelementptr inbounds { ptr, [[USIZE]] }, {{.*}}  %x, i32 0, i32 0
-// CHECK: store ptr %s.0, {{.*}} [[X0]]
-// CHECK: [[X1:%[0-9]+]] = getelementptr inbounds { ptr, [[USIZE]] }, {{.*}} %x, i32 0, i32 1
+// CHECK: store ptr %s.0, {{.*}} %x
+// CHECK: [[X1:%[0-9]+]] = getelementptr inbounds i8, {{.*}} %x, {{i32 4|i64 8}}
 // CHECK: store [[USIZE]] %s.1, {{.*}} [[X1]]
 
     let x = &*s;

tests/codegen/scalar-pair-bool.rs

@@ -2,25 +2,25 @@
 
 #![crate_type = "lib"]
 
-// CHECK: define{{.*}}{ i8, i8 } @pair_bool_bool(i1 noundef zeroext %pair.0, i1 noundef zeroext %pair.1)
+// CHECK: define{{.*}}{ i1, i1 } @pair_bool_bool(i1 noundef zeroext %pair.0, i1 noundef zeroext %pair.1)
 #[no_mangle]
 pub fn pair_bool_bool(pair: (bool, bool)) -> (bool, bool) {
     pair
 }
 
-// CHECK: define{{.*}}{ i8, i32 } @pair_bool_i32(i1 noundef zeroext %pair.0, i32 noundef %pair.1)
+// CHECK: define{{.*}}{ i1, i32 } @pair_bool_i32(i1 noundef zeroext %pair.0, i32 noundef %pair.1)
 #[no_mangle]
 pub fn pair_bool_i32(pair: (bool, i32)) -> (bool, i32) {
     pair
 }
 
-// CHECK: define{{.*}}{ i32, i8 } @pair_i32_bool(i32 noundef %pair.0, i1 noundef zeroext %pair.1)
+// CHECK: define{{.*}}{ i32, i1 } @pair_i32_bool(i32 noundef %pair.0, i1 noundef zeroext %pair.1)
 #[no_mangle]
 pub fn pair_i32_bool(pair: (i32, bool)) -> (i32, bool) {
     pair
 }
 
-// CHECK: define{{.*}}{ i8, i8 } @pair_and_or(i1 noundef zeroext %_1.0, i1 noundef zeroext %_1.1)
+// CHECK: define{{.*}}{ i1, i1 } @pair_and_or(i1 noundef zeroext %_1.0, i1 noundef zeroext %_1.1)
 #[no_mangle]
 pub fn pair_and_or((a, b): (bool, bool)) -> (bool, bool) {
     // Make sure it can operate directly on the unpacked args

tests/codegen/slice-iter-nonnull.rs

@@ -15,7 +15,7 @@
 // CHECK-LABEL: @slice_iter_next(
 #[no_mangle]
 pub fn slice_iter_next<'a>(it: &mut std::slice::Iter<'a, u32>) -> Option<&'a u32> {
-    // CHECK: %[[ENDP:.+]] = getelementptr{{.+}}ptr %it,{{.+}} 1
+    // CHECK: %[[ENDP:.+]] = getelementptr inbounds i8, ptr %it, {{i32 4|i64 8}}
     // CHECK: %[[END:.+]] = load ptr, ptr %[[ENDP]]
     // CHECK-SAME: !nonnull
     // CHECK-SAME: !noundef
@@ -32,7 +32,7 @@ pub fn slice_iter_next<'a>(it: &mut std::slice::Iter<'a, u32>) -> Option<&'a u32
 // CHECK-LABEL: @slice_iter_next_back(
 #[no_mangle]
 pub fn slice_iter_next_back<'a>(it: &mut std::slice::Iter<'a, u32>) -> Option<&'a u32> {
-    // CHECK: %[[ENDP:.+]] = getelementptr{{.+}}ptr %it,{{.+}} 1
+    // CHECK: %[[ENDP:.+]] = getelementptr inbounds i8, ptr %it, {{i32 4|i64 8}}
     // CHECK: %[[END:.+]] = load ptr, ptr %[[ENDP]]
     // CHECK-SAME: !nonnull
     // CHECK-SAME: !noundef
@@ -84,7 +84,7 @@ pub fn slice_iter_mut_new(slice: &mut [u32]) -> std::slice::IterMut<'_, u32> {
 // CHECK-LABEL: @slice_iter_is_empty
 #[no_mangle]
 pub fn slice_iter_is_empty(it: &std::slice::Iter<'_, u32>) -> bool {
-    // CHECK: %[[ENDP:.+]] = getelementptr{{.+}}ptr %it,{{.+}} 1
+    // CHECK: %[[ENDP:.+]] = getelementptr inbounds i8, ptr %it, {{i32 4|i64 8}}
     // CHECK: %[[END:.+]] = load ptr, ptr %[[ENDP]]
     // CHECK-SAME: !nonnull
     // CHECK-SAME: !noundef
@@ -100,7 +100,7 @@ pub fn slice_iter_is_empty(it: &std::slice::Iter<'_, u32>) -> bool {
 // CHECK-LABEL: @slice_iter_len
 #[no_mangle]
 pub fn slice_iter_len(it: &std::slice::Iter<'_, u32>) -> usize {
-    // CHECK: %[[ENDP:.+]] = getelementptr{{.+}}ptr %it,{{.+}} 1
+    // CHECK: %[[ENDP:.+]] = getelementptr inbounds i8, ptr %it, {{i32 4|i64 8}}
     // CHECK: %[[END:.+]] = load ptr, ptr %[[ENDP]]
     // CHECK-SAME: !nonnull
     // CHECK-SAME: !noundef