diff --git a/build_system/tests.rs b/build_system/tests.rs
index 1e24d1b113fe6..10736ff9a55c8 100644
--- a/build_system/tests.rs
+++ b/build_system/tests.rs
@@ -99,6 +99,7 @@ const BASE_SYSROOT_SUITE: &[TestCase] = &[
     TestCase::build_bin_and_run("aot.mod_bench", "example/mod_bench.rs", &[]),
     TestCase::build_bin_and_run("aot.issue-72793", "example/issue-72793.rs", &[]),
     TestCase::build_bin("aot.issue-59326", "example/issue-59326.rs"),
+    TestCase::build_bin_and_run("aot.neon", "example/neon.rs", &[]),
 ];
 
 pub(crate) static RAND_REPO: GitRepo = GitRepo::github(
diff --git a/config.txt b/config.txt
index 7ff805e58d967..2ccdc7d78748a 100644
--- a/config.txt
+++ b/config.txt
@@ -42,6 +42,7 @@ aot.float-minmax-pass
 aot.mod_bench
 aot.issue-72793
 aot.issue-59326
+aot.neon
 
 testsuite.extended_sysroot
 test.rust-random/rand
diff --git a/example/neon.rs b/example/neon.rs
new file mode 100644
index 0000000000000..bad26947967dc
--- /dev/null
+++ b/example/neon.rs
@@ -0,0 +1,234 @@
+// Most of these tests are copied from https://github.com/japaric/stdsimd/blob/0f4413d01c4f0c3ffbc5a69e9a37fbc7235b31a9/coresimd/arm/neon.rs
+
+#![feature(portable_simd)]
+
+#[cfg(target_arch = "aarch64")]
+use std::arch::aarch64::*;
+use std::mem::transmute;
+use std::simd::*;
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmin_s8() {
+    let a = i8x8::from([1, -2, 3, -4, 5, 6, 7, 8]);
+    let b = i8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
+    let e = i8x8::from([-2, -4, 5, 7, 0, 2, 4, 6]);
+    let r: i8x8 = transmute(vpmin_s8(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmin_s16() {
+    let a = i16x4::from([1, 2, 3, -4]);
+    let b = i16x4::from([0, 3, 2, 5]);
+    let e = i16x4::from([1, -4, 0, 2]);
+    let r: i16x4 = transmute(vpmin_s16(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmin_s32() {
+    let a = i32x2::from([1, -2]);
+    let b = i32x2::from([0, 3]);
+    let e = i32x2::from([-2, 0]);
+    let r: i32x2 = transmute(vpmin_s32(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmin_u8() {
+    let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 8]);
+    let b = u8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
+    let e = u8x8::from([1, 3, 5, 7, 0, 2, 4, 6]);
+    let r: u8x8 = transmute(vpmin_u8(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmin_u16() {
+    let a = u16x4::from([1, 2, 3, 4]);
+    let b = u16x4::from([0, 3, 2, 5]);
+    let e = u16x4::from([1, 3, 0, 2]);
+    let r: u16x4 = transmute(vpmin_u16(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmin_u32() {
+    let a = u32x2::from([1, 2]);
+    let b = u32x2::from([0, 3]);
+    let e = u32x2::from([1, 0]);
+    let r: u32x2 = transmute(vpmin_u32(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmin_f32() {
+    let a = f32x2::from([1., -2.]);
+    let b = f32x2::from([0., 3.]);
+    let e = f32x2::from([-2., 0.]);
+    let r: f32x2 = transmute(vpmin_f32(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmax_s8() {
+    let a = i8x8::from([1, -2, 3, -4, 5, 6, 7, 8]);
+    let b = i8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
+    let e = i8x8::from([1, 3, 6, 8, 3, 5, 7, 9]);
+    let r: i8x8 = transmute(vpmax_s8(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmax_s16() {
+    let a = i16x4::from([1, 2, 3, -4]);
+    let b = i16x4::from([0, 3, 2, 5]);
+    let e = i16x4::from([2, 3, 3, 5]);
+    let r: i16x4 = transmute(vpmax_s16(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmax_s32() {
+    let a = i32x2::from([1, -2]);
+    let b = i32x2::from([0, 3]);
+    let e = i32x2::from([1, 3]);
+    let r: i32x2 = transmute(vpmax_s32(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmax_u8() {
+    let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 8]);
+    let b = u8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
+    let e = u8x8::from([2, 4, 6, 8, 3, 5, 7, 9]);
+    let r: u8x8 = transmute(vpmax_u8(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmax_u16() {
+    let a = u16x4::from([1, 2, 3, 4]);
+    let b = u16x4::from([0, 3, 2, 5]);
+    let e = u16x4::from([2, 4, 3, 5]);
+    let r: u16x4 = transmute(vpmax_u16(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmax_u32() {
+    let a = u32x2::from([1, 2]);
+    let b = u32x2::from([0, 3]);
+    let e = u32x2::from([2, 3]);
+    let r: u32x2 = transmute(vpmax_u32(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpmax_f32() {
+    let a = f32x2::from([1., -2.]);
+    let b = f32x2::from([0., 3.]);
+    let e = f32x2::from([1., 3.]);
+    let r: f32x2 = transmute(vpmax_f32(transmute(a), transmute(b)));
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpadd_s16() {
+    let a = i16x4::from([1, 2, 3, 4]);
+    let b = i16x4::from([0, -1, -2, -3]);
+    let r: i16x4 = transmute(vpadd_s16(transmute(a), transmute(b)));
+    let e = i16x4::from([3, 7, -1, -5]);
+    assert_eq!(r, e);
+}
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpadd_s32() {
+    let a = i32x2::from([1, 2]);
+    let b = i32x2::from([0, -1]);
+    let r: i32x2 = transmute(vpadd_s32(transmute(a), transmute(b)));
+    let e = i32x2::from([3, -1]);
+    assert_eq!(r, e);
+}
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpadd_s8() {
+    let a = i8x8::from([1, 2, 3, 4, 5, 6, 7, 8]);
+    let b = i8x8::from([0, -1, -2, -3, -4, -5, -6, -7]);
+    let r: i8x8 = transmute(vpadd_s8(transmute(a), transmute(b)));
+    let e = i8x8::from([3, 7, 11, 15, -1, -5, -9, -13]);
+    assert_eq!(r, e);
+}
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpadd_u16() {
+    let a = u16x4::from([1, 2, 3, 4]);
+    let b = u16x4::from([30, 31, 32, 33]);
+    let r: u16x4 = transmute(vpadd_u16(transmute(a), transmute(b)));
+    let e = u16x4::from([3, 7, 61, 65]);
+    assert_eq!(r, e);
+}
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpadd_u32() {
+    let a = u32x2::from([1, 2]);
+    let b = u32x2::from([30, 31]);
+    let r: u32x2 = transmute(vpadd_u32(transmute(a), transmute(b)));
+    let e = u32x2::from([3, 61]);
+    assert_eq!(r, e);
+}
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vpadd_u8() {
+    let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 8]);
+    let b = u8x8::from([30, 31, 32, 33, 34, 35, 36, 37]);
+    let r: u8x8 = transmute(vpadd_u8(transmute(a), transmute(b)));
+    let e = u8x8::from([3, 7, 11, 15, 61, 65, 69, 73]);
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vqsub_u8() {
+    let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 0xff]);
+    let b = u8x8::from([30, 1, 1, 1, 34, 0xff, 36, 37]);
+    let r: u8x8 = transmute(vqsub_u8(transmute(a), transmute(b)));
+    let e = u8x8::from([0, 1, 2, 3, 0, 0, 0, 218]);
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+unsafe fn test_vqadd_u8() {
+    let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 0xff]);
+    let b = u8x8::from([30, 1, 1, 1, 34, 0xff, 36, 37]);
+    let r: u8x8 = transmute(vqadd_u8(transmute(a), transmute(b)));
+    let e = u8x8::from([31, 3, 4, 5, 39, 0xff, 43, 0xff]);
+    assert_eq!(r, e);
+}
+
+#[cfg(target_arch = "aarch64")]
+fn main() {
+    unsafe {
+        test_vpmin_s8();
+        test_vpmin_s16();
+        test_vpmin_s32();
+        test_vpmin_u8();
+        test_vpmin_u16();
+        test_vpmin_u32();
+        test_vpmin_f32();
+        test_vpmax_s8();
+        test_vpmax_s16();
+        test_vpmax_s32();
+        test_vpmax_u8();
+        test_vpmax_u16();
+        test_vpmax_u32();
+        test_vpmax_f32();
+
+        test_vpadd_s16();
+        test_vpadd_s32();
+        test_vpadd_s8();
+        test_vpadd_u16();
+        test_vpadd_u32();
+        test_vpadd_u8();
+
+        test_vqsub_u8();
+        test_vqadd_u8();
+    }
+}
+
+#[cfg(not(target_arch = "aarch64"))]
+fn main() {}
diff --git a/src/intrinsics/llvm_aarch64.rs b/src/intrinsics/llvm_aarch64.rs
index 0c211a06dc4a7..ee098be1fce6b 100644
--- a/src/intrinsics/llvm_aarch64.rs
+++ b/src/intrinsics/llvm_aarch64.rs
@@ -44,7 +44,9 @@ pub(crate) fn codegen_aarch64_llvm_intrinsic_call<'tcx>(
             });
         }
 
-        _ if intrinsic.starts_with("llvm.aarch64.neon.sqadd.v") => {
+        _ if intrinsic.starts_with("llvm.aarch64.neon.sqadd.v")
+            || intrinsic.starts_with("llvm.aarch64.neon.uqadd.v") =>
+        {
             intrinsic_args!(fx, args => (x, y); intrinsic);
 
             simd_pair_for_each_lane_typed(fx, x, y, ret, &|fx, x_lane, y_lane| {
@@ -52,7 +54,9 @@ pub(crate) fn codegen_aarch64_llvm_intrinsic_call<'tcx>(
             });
         }
 
-        _ if intrinsic.starts_with("llvm.aarch64.neon.sqsub.v") => {
+        _ if intrinsic.starts_with("llvm.aarch64.neon.sqsub.v")
+            || intrinsic.starts_with("llvm.aarch64.neon.uqsub.v") =>
+        {
             intrinsic_args!(fx, args => (x, y); intrinsic);
 
             simd_pair_for_each_lane_typed(fx, x, y, ret, &|fx, x_lane, y_lane| {
@@ -156,6 +160,90 @@ pub(crate) fn codegen_aarch64_llvm_intrinsic_call<'tcx>(
             });
         }
 
+        _ if intrinsic.starts_with("llvm.aarch64.neon.umaxp.v") => {
+            intrinsic_args!(fx, args => (x, y); intrinsic);
+
+            simd_horizontal_pair_for_each_lane(
+                fx,
+                x,
+                y,
+                ret,
+                &|fx, _lane_ty, _res_lane_ty, x_lane, y_lane| fx.bcx.ins().umax(x_lane, y_lane),
+            );
+        }
+
+        _ if intrinsic.starts_with("llvm.aarch64.neon.smaxp.v") => {
+            intrinsic_args!(fx, args => (x, y); intrinsic);
+
+            simd_horizontal_pair_for_each_lane(
+                fx,
+                x,
+                y,
+                ret,
+                &|fx, _lane_ty, _res_lane_ty, x_lane, y_lane| fx.bcx.ins().smax(x_lane, y_lane),
+            );
+        }
+
+        _ if intrinsic.starts_with("llvm.aarch64.neon.uminp.v") => {
+            intrinsic_args!(fx, args => (x, y); intrinsic);
+
+            simd_horizontal_pair_for_each_lane(
+                fx,
+                x,
+                y,
+                ret,
+                &|fx, _lane_ty, _res_lane_ty, x_lane, y_lane| fx.bcx.ins().umin(x_lane, y_lane),
+            );
+        }
+
+        _ if intrinsic.starts_with("llvm.aarch64.neon.sminp.v") => {
+            intrinsic_args!(fx, args => (x, y); intrinsic);
+
+            simd_horizontal_pair_for_each_lane(
+                fx,
+                x,
+                y,
+                ret,
+                &|fx, _lane_ty, _res_lane_ty, x_lane, y_lane| fx.bcx.ins().smin(x_lane, y_lane),
+            );
+        }
+
+        _ if intrinsic.starts_with("llvm.aarch64.neon.fminp.v") => {
+            intrinsic_args!(fx, args => (x, y); intrinsic);
+
+            simd_horizontal_pair_for_each_lane(
+                fx,
+                x,
+                y,
+                ret,
+                &|fx, _lane_ty, _res_lane_ty, x_lane, y_lane| fx.bcx.ins().fmin(x_lane, y_lane),
+            );
+        }
+
+        _ if intrinsic.starts_with("llvm.aarch64.neon.fmaxp.v") => {
+            intrinsic_args!(fx, args => (x, y); intrinsic);
+
+            simd_horizontal_pair_for_each_lane(
+                fx,
+                x,
+                y,
+                ret,
+                &|fx, _lane_ty, _res_lane_ty, x_lane, y_lane| fx.bcx.ins().fmax(x_lane, y_lane),
+            );
+        }
+
+        _ if intrinsic.starts_with("llvm.aarch64.neon.addp.v") => {
+            intrinsic_args!(fx, args => (x, y); intrinsic);
+
+            simd_horizontal_pair_for_each_lane(
+                fx,
+                x,
+                y,
+                ret,
+                &|fx, _lane_ty, _res_lane_ty, x_lane, y_lane| fx.bcx.ins().iadd(x_lane, y_lane),
+            );
+        }
+
         // FIXME generalize vector types
         "llvm.aarch64.neon.tbl1.v16i8" => {
             intrinsic_args!(fx, args => (t, idx); intrinsic);
@@ -172,25 +260,6 @@ pub(crate) fn codegen_aarch64_llvm_intrinsic_call<'tcx>(
             }
         }
 
-        // FIXME generalize vector types
-        "llvm.aarch64.neon.umaxp.v16i8" => {
-            intrinsic_args!(fx, args => (a, b); intrinsic);
-
-            // FIXME add helper for horizontal pairwise operations
-            for i in 0..8 {
-                let lane1 = a.value_lane(fx, i * 2).load_scalar(fx);
-                let lane2 = a.value_lane(fx, i * 2 + 1).load_scalar(fx);
-                let res = fx.bcx.ins().umax(lane1, lane2);
-                ret.place_lane(fx, i).to_ptr().store(fx, res, MemFlags::trusted());
-            }
-            for i in 0..8 {
-                let lane1 = b.value_lane(fx, i * 2).load_scalar(fx);
-                let lane2 = b.value_lane(fx, i * 2 + 1).load_scalar(fx);
-                let res = fx.bcx.ins().umax(lane1, lane2);
-                ret.place_lane(fx, 8 + i).to_ptr().store(fx, res, MemFlags::trusted());
-            }
-        }
-
         /*
         _ if intrinsic.starts_with("llvm.aarch64.neon.sshl.v")
             || intrinsic.starts_with("llvm.aarch64.neon.sqshl.v")
diff --git a/src/intrinsics/mod.rs b/src/intrinsics/mod.rs
index 58b31a534328e..bfeeb117ff5b3 100644
--- a/src/intrinsics/mod.rs
+++ b/src/intrinsics/mod.rs
@@ -132,6 +132,36 @@ fn simd_pair_for_each_lane<'tcx>(
     }
 }
 
+fn simd_horizontal_pair_for_each_lane<'tcx>(
+    fx: &mut FunctionCx<'_, '_, 'tcx>,
+    x: CValue<'tcx>,
+    y: CValue<'tcx>,
+    ret: CPlace<'tcx>,
+    f: &dyn Fn(&mut FunctionCx<'_, '_, 'tcx>, Ty<'tcx>, Ty<'tcx>, Value, Value) -> Value,
+) {
+    assert_eq!(x.layout(), y.layout());
+    let layout = x.layout();
+
+    let (lane_count, lane_ty) = layout.ty.simd_size_and_type(fx.tcx);
+    let lane_layout = fx.layout_of(lane_ty);
+    let (ret_lane_count, ret_lane_ty) = ret.layout().ty.simd_size_and_type(fx.tcx);
+    let ret_lane_layout = fx.layout_of(ret_lane_ty);
+    assert_eq!(lane_count, ret_lane_count);
+
+    for lane_idx in 0..lane_count {
+        let src = if lane_idx < (lane_count / 2) { x } else { y };
+        let src_idx = lane_idx % (lane_count / 2);
+
+        let lhs_lane = src.value_lane(fx, src_idx * 2).load_scalar(fx);
+        let rhs_lane = src.value_lane(fx, src_idx * 2 + 1).load_scalar(fx);
+
+        let res_lane = f(fx, lane_layout.ty, ret_lane_layout.ty, lhs_lane, rhs_lane);
+        let res_lane = CValue::by_val(res_lane, ret_lane_layout);
+
+        ret.place_lane(fx, lane_idx).write_cvalue(fx, res_lane);
+    }
+}
+
 fn simd_trio_for_each_lane<'tcx>(
     fx: &mut FunctionCx<'_, '_, 'tcx>,
     x: CValue<'tcx>,