Rollup of 4 pull requests

compiler/rustc_codegen_llvm/src/intrinsic.rs

@@ -2066,14 +2066,14 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
         };
     }
 
-    arith_red!(simd_reduce_add_ordered: vector_reduce_add, vector_reduce_fadd, true, add, 0.0);
+    arith_red!(simd_reduce_add_ordered: vector_reduce_add, vector_reduce_fadd, true, add, -0.0);
     arith_red!(simd_reduce_mul_ordered: vector_reduce_mul, vector_reduce_fmul, true, mul, 1.0);
     arith_red!(
         simd_reduce_add_unordered: vector_reduce_add,
         vector_reduce_fadd_reassoc,
         false,
         add,
-        0.0
+        -0.0
     );
     arith_red!(
         simd_reduce_mul_unordered: vector_reduce_mul,

compiler/rustc_target/src/spec/targets/armv8r_none_eabihf.rs

@@ -21,16 +21,16 @@ pub(crate) fn target() -> Target {
             linker: Some("rust-lld".into()),
             relocation_model: RelocModel::Static,
             panic_strategy: PanicStrategy::Abort,
-            // The Cortex-R52 has two variants with respect to floating-point support:
-            // 1. fp-armv8, SP-only, with 16 DP (32 SP) registers
-            // 2. neon-fp-armv8, SP+DP, with 32 DP registers
-            // Use the lesser of these two options as the default, as it will produce code
-            // compatible with either variant.
+            // Armv8-R requires a minimum set of floating-point features equivalent to:
+            // fp-armv8, SP-only, with 16 DP (32 SP) registers
+            // LLVM defines Armv8-R to include these features automatically.
+            //
+            // The Cortex-R52 supports these default features and optionally includes:
+            // neon-fp-armv8, SP+DP, with 32 DP registers
             //
             // Reference:
             // Arm Cortex-R52 Processor Technical Reference Manual
             // - Chapter 15 Advanced SIMD and floating-point support
-            features: "+fp-armv8,-fp64,-d32".into(),
             max_atomic_width: Some(64),
             emit_debug_gdb_scripts: false,
             // GCC defaults to 8 for arm-none here.

compiler/rustc_transmute/src/layout/tree.rs

@@ -175,7 +175,7 @@ pub(crate) mod rustc {
     use rustc_middle::ty::{self, AdtDef, AdtKind, List, ScalarInt, Ty, TyCtxt, TypeVisitableExt};
     use rustc_span::ErrorGuaranteed;
     use rustc_target::abi::{
-        FieldIdx, FieldsShape, Layout, Size, TyAndLayout, VariantIdx, Variants,
+        FieldIdx, FieldsShape, Layout, Size, TagEncoding, TyAndLayout, VariantIdx, Variants,
     };
 
     use super::Tree;
@@ -319,11 +319,17 @@ pub(crate) mod rustc {
             assert!(def.is_enum());
 
             // Computes the variant of a given index.
-            let layout_of_variant = |index| {
+            let layout_of_variant = |index, encoding: Option<TagEncoding<VariantIdx>>| {
                 let tag = cx.tcx.tag_for_variant((cx.tcx.erase_regions(ty), index));
                 let variant_def = Def::Variant(def.variant(index));
                 let variant_layout = ty_variant(cx, (ty, layout), index);
-                Self::from_variant(variant_def, tag, (ty, variant_layout), layout.size, cx)
+                Self::from_variant(
+                    variant_def,
+                    tag.map(|tag| (tag, index, encoding.unwrap())),
+                    (ty, variant_layout),
+                    layout.size,
+                    cx,
+                )
             };
 
             // We consider three kinds of enums, each demanding a different
@@ -345,9 +351,9 @@ pub(crate) mod rustc {
                 Variants::Single { index } => {
                     // `Variants::Single` on enums with variants denotes that
                     // the enum delegates its layout to the variant at `index`.
-                    layout_of_variant(*index)
+                    layout_of_variant(*index, None)
                 }
-                Variants::Multiple { tag_field, .. } => {
+                Variants::Multiple { tag, tag_encoding, tag_field, .. } => {
                     // `Variants::Multiple` denotes an enum with multiple
                     // variants. The layout of such an enum is the disjunction
                     // of the layouts of its tagged variants.
@@ -359,7 +365,7 @@ pub(crate) mod rustc {
                     let variants = def.discriminants(cx.tcx()).try_fold(
                         Self::uninhabited(),
                         |variants, (idx, ref discriminant)| {
-                            let variant = layout_of_variant(idx)?;
+                            let variant = layout_of_variant(idx, Some(tag_encoding.clone()))?;
                             Result::<Self, Err>::Ok(variants.or(variant))
                         },
                     )?;
@@ -380,7 +386,7 @@ pub(crate) mod rustc {
         /// `0`.
         fn from_variant(
             def: Def<'tcx>,
-            tag: Option<ScalarInt>,
+            tag: Option<(ScalarInt, VariantIdx, TagEncoding<VariantIdx>)>,
             (ty, layout): (Ty<'tcx>, Layout<'tcx>),
             total_size: Size,
             cx: LayoutCx<'tcx, TyCtxt<'tcx>>,
@@ -400,9 +406,18 @@ pub(crate) mod rustc {
             let mut struct_tree = Self::def(def);
 
             // If a `tag` is provided, place it at the start of the layout.
-            if let Some(tag) = tag {
-                size += tag.size();
-                struct_tree = struct_tree.then(Self::from_tag(tag, cx.tcx));
+            if let Some((tag, index, encoding)) = &tag {
+                match encoding {
+                    TagEncoding::Direct => {
+                        size += tag.size();
+                    }
+                    TagEncoding::Niche { niche_variants, .. } => {
+                        if !niche_variants.contains(index) {
+                            size += tag.size();
+                        }
+                    }
+                }
+                struct_tree = struct_tree.then(Self::from_tag(*tag, cx.tcx));
             }
 
             // Append the fields, in memory order, to the layout.

tests/assembly/simd/reduce-fadd-unordered.rs

@@ -0,0 +1,36 @@
+//@ revisions: x86_64 x86_64-avx2 aarch64
+//@ assembly-output: emit-asm
+//@ compile-flags: --crate-type=lib -O
+//@[aarch64] only-aarch64
+//@[x86_64] only-x86_64
+//@[x86_64-avx2] only-x86_64
+//@[x86_64-avx2] compile-flags: -Ctarget-cpu=x86-64-v3
+#![feature(portable_simd)]
+#![feature(core_intrinsics)]
+use std::intrinsics::simd as intrinsics;
+use std::simd::*;
+// Regression test for https://github.com/rust-lang/rust/issues/130028
+// This intrinsic produces much worse code if you use +0.0 instead of -0.0 because
+// +0.0 isn't as easy to algebraically reassociate, even using LLVM's reassoc attribute!
+// It would emit about an extra fadd, depending on the architecture.
+
+// CHECK-LABEL: reduce_fadd_negative_zero
+pub unsafe fn reduce_fadd_negative_zero(v: f32x4) -> f32 {
+    // x86_64: addps
+    // x86_64-NEXT: movaps
+    // x86_64-NEXT: shufps
+    // x86_64-NEXT: addss
+    // x86_64-NOT: xorps
+
+    // x86_64-avx2: vaddps
+    // x86_64-avx2-NEXT: vmovshdup
+    // x86_64-avx2-NEXT: vaddss
+    // x86_64-avx2-NOT: vxorps
+
+    // aarch64: faddp
+    // aarch64-NEXT: faddp
+
+    // CHECK-NOT: {{f?}}add{{p?s*}}
+    // CHECK: ret
+    intrinsics::simd_reduce_add_unordered(v)
+}

tests/ui/transmutability/enums/niche_optimization.rs

@@ -154,3 +154,12 @@ fn no_niche() {
     assert::is_transmutable::<Pair<V1, MaybeUninit<u8>>, OptionLike>();
     assert::is_transmutable::<Pair<V2, MaybeUninit<u8>>, OptionLike>();
 }
+
+fn niche_fields() {
+    enum Kind {
+        A(bool, bool),
+        B(bool),
+    }
+
+    assert::is_maybe_transmutable::<u16, Kind>();
+}