From 457ef6d36f34f33d6fa431646e2274048323dc3d Mon Sep 17 00:00:00 2001
From: Niklas Mischkulnig <4586894+mischnic@users.noreply.github.com>
Date: Wed, 18 Sep 2024 16:31:15 +0200
Subject: [PATCH 1/2] Add missing methods

---
 Cargo.lock          |  2 +-
 simd/src/arm/mod.rs | 40 +++++++++++++++++++++++++++++++++++++---
 2 files changed, 38 insertions(+), 4 deletions(-)

diff --git a/Cargo.lock b/Cargo.lock
index d2163d82..c61aff6d 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -2990,7 +2990,7 @@ source = "git+https://github.com/servo/pathfinder#e4fcda0d5259d0acf902aee6de7d25
 
 [[package]]
 name = "pathfinder_simd"
-version = "0.5.2"
+version = "0.5.4"
 dependencies = [
  "rustc_version 0.4.0",
 ]
diff --git a/simd/src/arm/mod.rs b/simd/src/arm/mod.rs
index deed1b7e..9bc49ee7 100644
--- a/simd/src/arm/mod.rs
+++ b/simd/src/arm/mod.rs
@@ -10,9 +10,9 @@
 
 use std::arch::aarch64::{self, float32x2_t, float32x4_t, int32x2_t, int32x4_t};
 use std::arch::aarch64::{uint32x2_t, uint32x4_t};
-use std::intrinsics::simd::*;
 use std::f32;
 use std::fmt::{self, Debug, Formatter};
+use std::intrinsics::simd::*;
 use std::mem;
 use std::ops::{Add, BitAnd, BitOr, Div, Index, IndexMut, Mul, Not, Shr, Sub};
 
@@ -201,7 +201,6 @@ impl IndexMut<usize> for F32x2 {
     }
 }
 
-
 impl Add<F32x2> for F32x2 {
     type Output = F32x2;
     #[inline]
@@ -365,6 +364,11 @@ impl F32x4 {
         unsafe { F32x4(simd_shuffle4!(self.0, other.0, [2, 3, 6, 7])) }
     }
 
+    #[inline]
+    pub fn concat_wz_yx(self, other: F32x4) -> F32x4 {
+        unsafe { F32x4(simd_shuffle4!(self.0, other.0, [3, 2, 5, 4])) }
+    }
+
     // Conversions
 
     /// Converts these packed floats to integers via rounding.
@@ -832,13 +836,22 @@ impl BitOr<U32x2> for U32x2 {
     }
 }
 
-
 // Four 32-bit unsigned integers
 
 #[derive(Clone, Copy)]
 pub struct U32x4(pub uint32x4_t);
 
 impl U32x4 {
+    #[inline]
+    pub fn new(a: u32, b: u32, c: u32, d: u32) -> U32x4 {
+        unsafe { U32x4(mem::transmute([a, b, c, d])) }
+    }
+
+    #[inline]
+    pub fn splat(x: u32) -> U32x4 {
+        U32x4::new(x, x, x, x)
+    }
+
     /// Returns true if all four booleans in this vector are true.
     ///
     /// The result is *undefined* if all four values in this vector are not booleans. A boolean is
@@ -856,6 +869,20 @@ impl U32x4 {
     pub fn all_false(&self) -> bool {
         unsafe { aarch64::vmaxvq_u32(self.0) == 0 }
     }
+
+    // Packed comparisons
+
+    #[inline]
+    pub fn packed_eq(self, other: U32x4) -> U32x4 {
+        unsafe { U32x4(simd_eq(self.0, other.0)) }
+    }
+}
+
+impl Debug for U32x4 {
+    #[inline]
+    fn fmt(&self, f: &mut Formatter) -> Result<(), fmt::Error> {
+        write!(f, "<{}, {}, {}, {}>", self[0], self[1], self[2], self[3])
+    }
 }
 
 impl Index<usize> for U32x4 {
@@ -870,6 +897,13 @@ impl Index<usize> for U32x4 {
     }
 }
 
+impl PartialEq for U32x4 {
+    #[inline]
+    fn eq(&self, other: &U32x4) -> bool {
+        self.packed_eq(*other).all_true()
+    }
+}
+
 extern "C" {
     #[link_name = "llvm.fabs.v2f32"]
     fn fabs_v2f32(a: float32x2_t) -> float32x2_t;

From bead563cd373cf225fc3565ea4239a005463ebf0 Mon Sep 17 00:00:00 2001
From: Niklas Mischkulnig <4586894+mischnic@users.noreply.github.com>
Date: Wed, 18 Sep 2024 17:23:16 +0200
Subject: [PATCH 2/2] Fix arm F32x4::concat_xy_xy

---
 simd/src/arm/mod.rs | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/simd/src/arm/mod.rs b/simd/src/arm/mod.rs
index 9bc49ee7..54a59ff1 100644
--- a/simd/src/arm/mod.rs
+++ b/simd/src/arm/mod.rs
@@ -351,7 +351,7 @@ impl F32x4 {
 
     #[inline]
     pub fn concat_xy_xy(self, other: F32x4) -> F32x4 {
-        unsafe { F32x4(simd_shuffle4!(self.0, other.0, [0, 1, 2, 3])) }
+        unsafe { F32x4(simd_shuffle4!(self.0, other.0, [0, 1, 4, 5])) }
     }
 
     #[inline]