Add missing vtst_p16 and vtstq_p16 intrinsics (#1257)

Author: Amanieu

crates/core_arch/src/arm_shared/neon/generated.rs

@@ -1032,6 +1032,30 @@ pub unsafe fn vtstq_p8(a: poly8x16_t, b: poly8x16_t) -> uint8x16_t {
     simd_ne(c, transmute(d))
 }
 
+/// Signed compare bitwise Test bits nonzero
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
+pub unsafe fn vtst_p16(a: poly16x4_t, b: poly16x4_t) -> uint16x4_t {
+    let c: poly16x4_t = simd_and(a, b);
+    let d: i16x4 = i16x4::new(0, 0, 0, 0);
+    simd_ne(c, transmute(d))
+}
+
+/// Signed compare bitwise Test bits nonzero
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr(vtst))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(cmtst))]
+pub unsafe fn vtstq_p16(a: poly16x8_t, b: poly16x8_t) -> uint16x8_t {
+    let c: poly16x8_t = simd_and(a, b);
+    let d: i16x8 = i16x8::new(0, 0, 0, 0, 0, 0, 0, 0);
+    simd_ne(c, transmute(d))
+}
+
 /// Unsigned compare bitwise Test bits nonzero
 #[inline]
 #[target_feature(enable = "neon")]
@@ -25039,6 +25063,24 @@ mod test {
         assert_eq!(r, e);
     }
 
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vtst_p16() {
+        let a: i16x4 = i16x4::new(-32768, 0x00, 0x01, 0x02);
+        let b: i16x4 = i16x4::new(-32768, 0x00, 0x01, 0x02);
+        let e: u16x4 = u16x4::new(0xFF_FF, 0, 0xFF_FF, 0xFF_FF);
+        let r: u16x4 = transmute(vtst_p16(transmute(a), transmute(b)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vtstq_p16() {
+        let a: i16x8 = i16x8::new(-32768, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06);
+        let b: i16x8 = i16x8::new(-32768, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06);
+        let e: u16x8 = u16x8::new(0xFF_FF, 0, 0xFF_FF, 0xFF_FF, 0xFF_FF, 0xFF_FF, 0xFF_FF, 0xFF_FF);
+        let r: u16x8 = transmute(vtstq_p16(transmute(a), transmute(b)));
+        assert_eq!(r, e);
+    }
+
     #[simd_test(enable = "neon")]
     unsafe fn test_vtst_u8() {
         let a: u8x8 = u8x8::new(0, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06);

crates/stdarch-gen/neon.spec

@@ -411,7 +411,7 @@ aarch64 = cmtst
 generate int64x1_t:uint64x1_t, int64x2_t:uint64x2_t, poly64x1_t:uint64x1_t, poly64x2_t:uint64x2_t
 
 arm = vtst
-generate int8x8_t:uint8x8_t, int8x16_t:uint8x16_t, int16x4_t:uint16x4_t, int16x8_t:uint16x8_t, int32x2_t:uint32x2_t, int32x4_t:uint32x4_t, poly8x8_t:uint8x8_t, poly8x16_t:uint8x16_t
+generate int8x8_t:uint8x8_t, int8x16_t:uint8x16_t, int16x4_t:uint16x4_t, int16x8_t:uint16x8_t, int32x2_t:uint32x2_t, int32x4_t:uint32x4_t, poly8x8_t:uint8x8_t, poly8x16_t:uint8x16_t, poly16x4_t:uint16x4_t, poly16x8_t:uint16x8_t
 
 /// Unsigned compare bitwise Test bits nonzero
 name = vtst
@@ -7550,4 +7550,4 @@ n = 2
 validate 510
 
 aarch64 = sri
-generate i64, u64
+generate i64, u64

crates/stdarch-verify/tests/arm.rs

@@ -594,6 +594,8 @@ fn verify_all_signatures() {
             "vclsq_u8",
             "vclsq_u16",
             "vclsq_u32",
+            "vtst_p16",
+            "vtstq_p16",
             "__dbg",
         ];
         let arm = match map.get(rust.name) {