From 70a6abfd29c20fbaa7bee58259a16a62ed48ad93 Mon Sep 17 00:00:00 2001
From: Afonso Bordado <afonsobordado@az8.co>
Date: Sat, 4 Nov 2023 19:37:36 +0000
Subject: [PATCH] Add unsigned saturating add/sub intrinsics for aarch64

---
 example/neon.rs                | 21 +++++++++++++++++++++
 src/intrinsics/llvm_aarch64.rs |  8 ++++++--
 2 files changed, 27 insertions(+), 2 deletions(-)

diff --git a/example/neon.rs b/example/neon.rs
index 6ea053d0b00..10337137322 100644
--- a/example/neon.rs
+++ b/example/neon.rs
@@ -180,6 +180,24 @@ unsafe fn test_vpadd_u8() {
     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 {
@@ -204,6 +222,9 @@ fn main() {
         test_vpadd_u16();
         test_vpadd_u32();
         test_vpadd_u8();
+
+        test_vqsub_u8();
+        test_vqadd_u8();
     }
 }
 
diff --git a/src/intrinsics/llvm_aarch64.rs b/src/intrinsics/llvm_aarch64.rs
index fdad5a474d6..ee098be1fce 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| {