add neon instruction vaddlv_*

crates/core_arch/src/aarch64/neon/generated.rs

@@ -2876,6 +2876,110 @@ pub unsafe fn vsubq_f64(a: float64x2_t, b: float64x2_t) -> float64x2_t {
     simd_sub(a, b)
 }
 
+/// Signed Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(saddlv))]
+pub unsafe fn vaddlv_s16(a: int16x4_t) -> i32 {
+    #[allow(improper_ctypes)]
+    extern "C" {
+        #[cfg_attr(target_arch = "aarch64", link_name = "llvm.aarch64.neon.saddlv.i32.v4i16")]
+        fn vaddlv_s16_(a: int16x4_t) -> i32;
+    }
+    vaddlv_s16_(a)
+}
+
+/// Signed Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(saddlv))]
+pub unsafe fn vaddlvq_s16(a: int16x8_t) -> i32 {
+    #[allow(improper_ctypes)]
+    extern "C" {
+        #[cfg_attr(target_arch = "aarch64", link_name = "llvm.aarch64.neon.saddlv.i32.v8i16")]
+        fn vaddlvq_s16_(a: int16x8_t) -> i32;
+    }
+    vaddlvq_s16_(a)
+}
+
+/// Signed Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(saddlp))]
+pub unsafe fn vaddlv_s32(a: int32x2_t) -> i64 {
+    #[allow(improper_ctypes)]
+    extern "C" {
+        #[cfg_attr(target_arch = "aarch64", link_name = "llvm.aarch64.neon.saddlv.i64.v2i32")]
+        fn vaddlv_s32_(a: int32x2_t) -> i64;
+    }
+    vaddlv_s32_(a)
+}
+
+/// Signed Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(saddlv))]
+pub unsafe fn vaddlvq_s32(a: int32x4_t) -> i64 {
+    #[allow(improper_ctypes)]
+    extern "C" {
+        #[cfg_attr(target_arch = "aarch64", link_name = "llvm.aarch64.neon.saddlv.i64.v4i32")]
+        fn vaddlvq_s32_(a: int32x4_t) -> i64;
+    }
+    vaddlvq_s32_(a)
+}
+
+/// Unsigned Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(uaddlv))]
+pub unsafe fn vaddlv_u16(a: uint16x4_t) -> u32 {
+    #[allow(improper_ctypes)]
+    extern "C" {
+        #[cfg_attr(target_arch = "aarch64", link_name = "llvm.aarch64.neon.uaddlv.i32.v4i16")]
+        fn vaddlv_u16_(a: uint16x4_t) -> u32;
+    }
+    vaddlv_u16_(a)
+}
+
+/// Unsigned Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(uaddlv))]
+pub unsafe fn vaddlvq_u16(a: uint16x8_t) -> u32 {
+    #[allow(improper_ctypes)]
+    extern "C" {
+        #[cfg_attr(target_arch = "aarch64", link_name = "llvm.aarch64.neon.uaddlv.i32.v8i16")]
+        fn vaddlvq_u16_(a: uint16x8_t) -> u32;
+    }
+    vaddlvq_u16_(a)
+}
+
+/// Unsigned Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(uaddlp))]
+pub unsafe fn vaddlv_u32(a: uint32x2_t) -> u64 {
+    #[allow(improper_ctypes)]
+    extern "C" {
+        #[cfg_attr(target_arch = "aarch64", link_name = "llvm.aarch64.neon.uaddlv.i64.v2i32")]
+        fn vaddlv_u32_(a: uint32x2_t) -> u64;
+    }
+    vaddlv_u32_(a)
+}
+
+/// Unsigned Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(uaddlv))]
+pub unsafe fn vaddlvq_u32(a: uint32x4_t) -> u64 {
+    #[allow(improper_ctypes)]
+    extern "C" {
+        #[cfg_attr(target_arch = "aarch64", link_name = "llvm.aarch64.neon.uaddlv.i64.v4i32")]
+        fn vaddlvq_u32_(a: uint32x4_t) -> u64;
+    }
+    vaddlvq_u32_(a)
+}
+
 /// Signed Subtract Wide
 #[inline]
 #[target_feature(enable = "neon")]
@@ -8458,6 +8562,70 @@ mod test {
         assert_eq!(r, e);
     }
 
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlv_s16() {
+        let a: i16x4 = i16x4::new(1, 2, 3, 4);
+        let e: i32 = 10;
+        let r: i32 = transmute(vaddlv_s16(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlvq_s16() {
+        let a: i16x8 = i16x8::new(1, 2, 3, 4, 5, 6, 7, 8);
+        let e: i32 = 36;
+        let r: i32 = transmute(vaddlvq_s16(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlv_s32() {
+        let a: i32x2 = i32x2::new(1, 2);
+        let e: i64 = 3;
+        let r: i64 = transmute(vaddlv_s32(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlvq_s32() {
+        let a: i32x4 = i32x4::new(1, 2, 3, 4);
+        let e: i64 = 10;
+        let r: i64 = transmute(vaddlvq_s32(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlv_u16() {
+        let a: u16x4 = u16x4::new(1, 2, 3, 4);
+        let e: u32 = 10;
+        let r: u32 = transmute(vaddlv_u16(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlvq_u16() {
+        let a: u16x8 = u16x8::new(1, 2, 3, 4, 5, 6, 7, 8);
+        let e: u32 = 36;
+        let r: u32 = transmute(vaddlvq_u16(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlv_u32() {
+        let a: u32x2 = u32x2::new(1, 2);
+        let e: u64 = 3;
+        let r: u64 = transmute(vaddlv_u32(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlvq_u32() {
+        let a: u32x4 = u32x4::new(1, 2, 3, 4);
+        let e: u64 = 10;
+        let r: u64 = transmute(vaddlvq_u32(transmute(a)));
+        assert_eq!(r, e);
+    }
+
     #[simd_test(enable = "neon")]
     unsafe fn test_vsubw_high_s8() {
         let a: i16x8 = i16x8::new(8, 9, 10, 12, 13, 14, 15, 16);

crates/core_arch/src/aarch64/neon/mod.rs

@@ -147,6 +147,15 @@ extern "C" {
     #[link_name = "llvm.aarch64.neon.uaddv.i64.v2i64"]
     fn vaddvq_u64_(a: uint64x2_t) -> u64;
 
+    #[link_name = "llvm.aarch64.neon.saddlv.i32.v8i8"]
+    fn vaddlv_s8_(a: int8x8_t) -> i32;
+    #[link_name = "llvm.aarch64.neon.uaddlv.i32.v8i8"]
+    fn vaddlv_u8_(a: uint8x8_t) -> u32;
+    #[link_name = "llvm.aarch64.neon.saddlv.i32.v16i8"]
+    fn vaddlvq_s8_(a: int8x16_t) -> i32;
+    #[link_name = "llvm.aarch64.neon.uaddlv.i32.v16i8"]
+    fn vaddlvq_u8_(a: uint8x16_t) -> u32;
+
     #[link_name = "llvm.aarch64.neon.smaxv.i8.v8i8"]
     fn vmaxv_s8_(a: int8x8_t) -> i8;
     #[link_name = "llvm.aarch64.neon.smaxv.i8.6i8"]
@@ -1000,6 +1009,35 @@ pub unsafe fn vaddvq_u64(a: uint64x2_t) -> u64 {
     vaddvq_u64_(a)
 }
 
+/// Signed Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(saddlv))]
+pub unsafe fn vaddlv_s8(a: int8x8_t) -> i16 {
+    vaddlv_s8_(a) as i16
+}
+/// Signed Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(saddlv))]
+pub unsafe fn vaddlvq_s8(a: int8x16_t) -> i16 {
+    vaddlvq_s8_(a) as i16
+}
+/// Unsigned Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(uaddlv))]
+pub unsafe fn vaddlv_u8(a: uint8x8_t) -> u16 {
+    vaddlv_u8_(a) as u16
+}
+/// Unsigned Add Long across Vector
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(uaddlv))]
+pub unsafe fn vaddlvq_u8(a: uint8x16_t) -> u16 {
+    vaddlvq_u8_(a) as u16
+}
+
 /// Polynomial multiply long
 #[inline]
 #[target_feature(enable = "neon")]
@@ -4367,6 +4405,35 @@ mod tests {
         let e = 3_u64;
         assert_eq!(r, e);
     }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlv_s8() {
+        let a = i8x8::new(1, 2, 3, 4, 5, 6, 7, -8);
+        let r: i16 = vaddlv_s8(transmute(a));
+        let e = 20_i16;
+        assert_eq!(r, e);
+    }
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlv_u8() {
+        let a = u8x8::new(1, 2, 3, 4, 5, 6, 7, 8);
+        let r: u16 = vaddlv_u8(transmute(a));
+        let e = 36_u16;
+        assert_eq!(r, e);
+    }
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlvq_s8() {
+        let a = i8x16::new(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, -16);
+        let r: i16 = vaddlvq_s8(transmute(a));
+        let e = 104_i16;
+        assert_eq!(r, e);
+    }
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vaddlvq_u8() {
+        let a = u8x16::new(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16);
+        let r: u16 = vaddlvq_u8(transmute(a));
+        let e = 136_u16;
+        assert_eq!(r, e);
+    }
 }
 
 #[cfg(test)]

crates/stdarch-gen/neon.spec

@@ -1655,6 +1655,78 @@ generate float64x*_t
 arm = vsub.
 generate float*_t
 
+/// Signed Add Long across Vector
+name = vaddlv
+a = 1, 2, 3, 4
+validate 10
+
+aarch64 = saddlv
+link-aarch64 = llvm.aarch64.neon.saddlv.i32._EXT_
+generate int16x4_t:i32
+
+/// Signed Add Long across Vector
+name = vaddlv
+a = 1, 2, 3, 4, 5, 6, 7, 8
+validate 36
+
+aarch64 = saddlv
+link-aarch64 = llvm.aarch64.neon.saddlv.i32._EXT_
+generate int16x8_t:i32
+
+/// Signed Add Long across Vector
+name = vaddlv
+a = 1, 2
+validate 3
+
+aarch64 = saddlp
+link-aarch64 = llvm.aarch64.neon.saddlv.i64._EXT_
+generate int32x2_t:i64
+
+/// Signed Add Long across Vector
+name = vaddlv
+a = 1, 2, 3, 4
+validate 10
+
+aarch64 = saddlv
+link-aarch64 = llvm.aarch64.neon.saddlv.i64._EXT_
+generate int32x4_t:i64
+
+/// Unsigned Add Long across Vector
+name = vaddlv
+a = 1, 2, 3, 4
+validate 10
+
+aarch64 = uaddlv
+link-aarch64 = llvm.aarch64.neon.uaddlv.i32._EXT_
+generate uint16x4_t:u32
+
+/// Unsigned Add Long across Vector
+name = vaddlv
+a = 1, 2, 3, 4, 5, 6, 7, 8
+validate 36
+
+aarch64 = uaddlv
+link-aarch64 = llvm.aarch64.neon.uaddlv.i32._EXT_
+generate uint16x8_t:u32
+
+/// Unsigned Add Long across Vector
+name = vaddlv
+a = 1, 2
+validate 3
+
+aarch64 = uaddlp
+link-aarch64 = llvm.aarch64.neon.uaddlv.i64._EXT_
+generate uint32x2_t:u64
+
+/// Unsigned Add Long across Vector
+name = vaddlv
+a = 1, 2, 3, 4
+validate 10
+
+aarch64 = uaddlv
+link-aarch64 = llvm.aarch64.neon.uaddlv.i64._EXT_
+generate uint32x4_t:u64
+
 /// Subtract returning high narrow
 name = vsubhn
 no-q