add neon instruction vget_low_* and fix vget_high_* (#1082)

Author: surechen

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

@@ -1520,6 +1520,30 @@ pub unsafe fn vget_high_f64(a: float64x2_t) -> float64x1_t {
     float64x1_t(simd_extract(a, 1))
 }
 
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(ldr))]
+pub unsafe fn vget_high_p64(a: poly64x2_t) -> poly64x1_t {
+    transmute(u64x1::new(simd_extract(a, 1)))
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(ldr))]
+pub unsafe fn vget_low_f64(a: float64x2_t) -> float64x1_t {
+    float64x1_t(simd_extract(a, 0))
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(test, assert_instr(ldr))]
+pub unsafe fn vget_low_p64(a: poly64x2_t) -> poly64x1_t {
+    transmute(u64x1::new(simd_extract(a, 0)))
+}
+
 /* FIXME: 16-bit float
 /// Vector combine
 #[inline]
@@ -3483,6 +3507,30 @@ mod tests {
         assert_eq!(r, e);
     }
 
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_high_p64() {
+        let a = u64x2::new(1, 2);
+        let e = u64x1::new(2);
+        let r: u64x1 = transmute(vget_high_p64(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_f64() {
+        let a = f64x2::new(1.0, 2.0);
+        let e = f64x1::new(1.0);
+        let r: f64x1 = transmute(vget_low_f64(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_p64() {
+        let a = u64x2::new(1, 2);
+        let e = u64x1::new(1);
+        let r: u64x1 = transmute(vget_low_p64(transmute(a)));
+        assert_eq!(r, e);
+    }
+
     #[simd_test(enable = "neon")]
     unsafe fn test_vceq_u64() {
         test_cmp_u64(

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

@@ -3848,18 +3848,118 @@ pub unsafe fn vget_high_p16(a: poly16x8_t) -> poly16x4_t {
 #[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
 #[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmov"))]
 #[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext))]
-pub unsafe fn vget_high_p64(a: poly64x2_t) -> poly64x1_t {
-    poly64x1_t(simd_extract(a, 1))
+pub unsafe fn vget_high_f32(a: float32x4_t) -> float32x2_t {
+    simd_shuffle2(a, a, [2, 3])
 }
 
 /// Duplicate vector element to vector or scalar
 #[inline]
 #[target_feature(enable = "neon")]
 #[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
-#[cfg_attr(all(test, target_arch = "arm"), assert_instr("vmov"))]
-#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ext))]
-pub unsafe fn vget_high_f32(a: float32x4_t) -> float32x2_t {
-    simd_shuffle2(a, a, [2, 3])
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr("ldr"))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ldr))]
+pub unsafe fn vget_low_s8(a: int8x16_t) -> int8x8_t {
+    simd_shuffle8(a, a, [0, 1, 2, 3, 4, 5, 6, 7])
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr("ldr"))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ldr))]
+pub unsafe fn vget_low_s16(a: int16x8_t) -> int16x4_t {
+    simd_shuffle4(a, a, [0, 1, 2, 3])
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr("ldr"))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ldr))]
+pub unsafe fn vget_low_s32(a: int32x4_t) -> int32x2_t {
+    simd_shuffle2(a, a, [0, 1])
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr("ldr"))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ldr))]
+pub unsafe fn vget_low_s64(a: int64x2_t) -> int64x1_t {
+    int64x1_t(simd_extract(a, 0))
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr("ldr"))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ldr))]
+pub unsafe fn vget_low_u8(a: uint8x16_t) -> uint8x8_t {
+    simd_shuffle8(a, a, [0, 1, 2, 3, 4, 5, 6, 7])
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr("ldr"))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ldr))]
+pub unsafe fn vget_low_u16(a: uint16x8_t) -> uint16x4_t {
+    simd_shuffle4(a, a, [0, 1, 2, 3])
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr("ldr"))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ldr))]
+pub unsafe fn vget_low_u32(a: uint32x4_t) -> uint32x2_t {
+    simd_shuffle2(a, a, [0, 1])
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr("ldr"))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ldr))]
+pub unsafe fn vget_low_u64(a: uint64x2_t) -> uint64x1_t {
+    uint64x1_t(simd_extract(a, 0))
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr("ldr"))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ldr))]
+pub unsafe fn vget_low_p8(a: poly8x16_t) -> poly8x8_t {
+    simd_shuffle8(a, a, [0, 1, 2, 3, 4, 5, 6, 7])
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr("ldr"))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ldr))]
+pub unsafe fn vget_low_p16(a: poly16x8_t) -> poly16x4_t {
+    simd_shuffle4(a, a, [0, 1, 2, 3])
+}
+
+/// Duplicate vector element to vector or scalar
+#[inline]
+#[target_feature(enable = "neon")]
+#[cfg_attr(target_arch = "arm", target_feature(enable = "v7"))]
+#[cfg_attr(all(test, target_arch = "arm"), assert_instr("ldr"))]
+#[cfg_attr(all(test, target_arch = "aarch64"), assert_instr(ldr))]
+pub unsafe fn vget_low_f32(a: float32x4_t) -> float32x2_t {
+    simd_shuffle2(a, a, [0, 1])
 }
 
 /// Duplicate vector element to vector or scalar
@@ -5853,31 +5953,31 @@ mod tests {
     unsafe fn test_vget_high_u8() {
         let a = u8x16::new(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16);
         let e = u8x8::new(9, 10, 11, 12, 13, 14, 15, 16);
-        let r: u8x8 = transmute(vget_high_s8(transmute(a)));
+        let r: u8x8 = transmute(vget_high_u8(transmute(a)));
         assert_eq!(r, e);
     }
 
     #[simd_test(enable = "neon")]
     unsafe fn test_vget_high_u16() {
         let a = u16x8::new(1, 2, 3, 4, 5, 6, 7, 8);
         let e = u16x4::new(5, 6, 7, 8);
-        let r: u16x4 = transmute(vget_high_s16(transmute(a)));
+        let r: u16x4 = transmute(vget_high_u16(transmute(a)));
         assert_eq!(r, e);
     }
 
     #[simd_test(enable = "neon")]
     unsafe fn test_vget_high_u32() {
         let a = u32x4::new(1, 2, 3, 4);
         let e = u32x2::new(3, 4);
-        let r: u32x2 = transmute(vget_high_s32(transmute(a)));
+        let r: u32x2 = transmute(vget_high_u32(transmute(a)));
         assert_eq!(r, e);
     }
 
     #[simd_test(enable = "neon")]
     unsafe fn test_vget_high_u64() {
         let a = u64x2::new(1, 2);
         let e = u64x1::new(2);
-        let r: u64x1 = transmute(vget_high_s64(transmute(a)));
+        let r: u64x1 = transmute(vget_high_u64(transmute(a)));
         assert_eq!(r, e);
     }
 
@@ -5898,18 +5998,98 @@ mod tests {
     }
 
     #[simd_test(enable = "neon")]
-    unsafe fn test_vget_high_p64() {
+    unsafe fn test_vget_high_f32() {
+        let a = f32x4::new(1.0, 2.0, 3.0, 4.0);
+        let e = f32x2::new(3.0, 4.0);
+        let r: f32x2 = transmute(vget_high_f32(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_s8() {
+        let a = i8x16::new(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16);
+        let e = i8x8::new(1, 2, 3, 4, 5, 6, 7, 8);
+        let r: i8x8 = transmute(vget_low_s8(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_s16() {
+        let a = i16x8::new(1, 2, 3, 4, 5, 6, 7, 8);
+        let e = i16x4::new(1, 2, 3, 4);
+        let r: i16x4 = transmute(vget_low_s16(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_s32() {
+        let a = i32x4::new(1, 2, 3, 4);
+        let e = i32x2::new(1, 2);
+        let r: i32x2 = transmute(vget_low_s32(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_s64() {
+        let a = i64x2::new(1, 2);
+        let e = i64x1::new(1);
+        let r: i64x1 = transmute(vget_low_s64(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_u8() {
+        let a = u8x16::new(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16);
+        let e = u8x8::new(1, 2, 3, 4, 5, 6, 7, 8);
+        let r: u8x8 = transmute(vget_low_u8(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_u16() {
+        let a = u16x8::new(1, 2, 3, 4, 5, 6, 7, 8);
+        let e = u16x4::new(1, 2, 3, 4);
+        let r: u16x4 = transmute(vget_low_u16(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_u32() {
+        let a = u32x4::new(1, 2, 3, 4);
+        let e = u32x2::new(1, 2);
+        let r: u32x2 = transmute(vget_low_u32(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_u64() {
         let a = u64x2::new(1, 2);
-        let e = u64x1::new(2);
-        let r: u64x1 = transmute(vget_high_p64(transmute(a)));
+        let e = u64x1::new(1);
+        let r: u64x1 = transmute(vget_low_u64(transmute(a)));
         assert_eq!(r, e);
     }
 
     #[simd_test(enable = "neon")]
-    unsafe fn test_vget_high_f32() {
+    unsafe fn test_vget_low_p8() {
+        let a = u8x16::new(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16);
+        let e = u8x8::new(1, 2, 3, 4, 5, 6, 7, 8);
+        let r: u8x8 = transmute(vget_low_p8(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_p16() {
+        let a = u16x8::new(1, 2, 3, 4, 5, 6, 7, 8);
+        let e = u16x4::new(1, 2, 3, 4);
+        let r: u16x4 = transmute(vget_low_p16(transmute(a)));
+        assert_eq!(r, e);
+    }
+
+    #[simd_test(enable = "neon")]
+    unsafe fn test_vget_low_f32() {
         let a = f32x4::new(1.0, 2.0, 3.0, 4.0);
-        let e = f32x2::new(3.0, 4.0);
-        let r: f32x2 = transmute(vget_high_f32(transmute(a)));
+        let e = f32x2::new(1.0, 2.0);
+        let r: f32x2 = transmute(vget_low_f32(transmute(a)));
         assert_eq!(r, e);
     }