aarch64: implement vabs using instrinsics::simd

Author: folkertdev

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

@@ -298,46 +298,40 @@ pub fn vabsq_f64(a: float64x2_t) -> float64x2_t {
 #[stable(feature = "neon_intrinsics", since = "1.59.0")]
 #[cfg_attr(test, assert_instr(abs))]
 pub fn vabs_s64(a: int64x1_t) -> int64x1_t {
-    unsafe extern "unadjusted" {
-        #[cfg_attr(
-            any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.abs.v1i64"
-        )]
-        fn _vabs_s64(a: int64x1_t) -> int64x1_t;
+    unsafe {
+        let neg: int64x1_t = simd_neg(a);
+        let mask: int64x1_t = simd_ge(a, neg);
+        simd_select(mask, a, neg)
     }
-    unsafe { _vabs_s64(a) }
 }
 #[doc = "Absolute Value (wrapping)."]
-#[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vabsd_s64)"]
+#[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vabsq_s64)"]
 #[inline]
 #[target_feature(enable = "neon")]
 #[stable(feature = "neon_intrinsics", since = "1.59.0")]
 #[cfg_attr(test, assert_instr(abs))]
-pub fn vabsd_s64(a: i64) -> i64 {
-    unsafe extern "unadjusted" {
-        #[cfg_attr(
-            any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.abs.i64"
-        )]
-        fn _vabsd_s64(a: i64) -> i64;
+pub fn vabsq_s64(a: int64x2_t) -> int64x2_t {
+    unsafe {
+        let neg: int64x2_t = simd_neg(a);
+        let mask: int64x2_t = simd_ge(a, neg);
+        simd_select(mask, a, neg)
     }
-    unsafe { _vabsd_s64(a) }
 }
 #[doc = "Absolute Value (wrapping)."]
-#[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vabsq_s64)"]
+#[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vabsd_s64)"]
 #[inline]
 #[target_feature(enable = "neon")]
 #[stable(feature = "neon_intrinsics", since = "1.59.0")]
 #[cfg_attr(test, assert_instr(abs))]
-pub fn vabsq_s64(a: int64x2_t) -> int64x2_t {
+pub fn vabsd_s64(a: i64) -> i64 {
     unsafe extern "unadjusted" {
         #[cfg_attr(
             any(target_arch = "aarch64", target_arch = "arm64ec"),
-            link_name = "llvm.aarch64.neon.abs.v2i64"
+            link_name = "llvm.aarch64.neon.abs.i64"
         )]
-        fn _vabsq_s64(a: int64x2_t) -> int64x2_t;
+        fn _vabsd_s64(a: i64) -> i64;
     }
-    unsafe { _vabsq_s64(a) }
+    unsafe { _vabsd_s64(a) }
 }
 #[doc = "Add"]
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vaddd_s64)"]

crates/stdarch-gen-arm/spec/neon/aarch64.spec.yml

@@ -13023,6 +13023,25 @@ intrinsics:
             - link: "llvm.aarch64.crc32cx"
               arch: aarch64,arm64ec
 
+  - name: "vabsd_s64"
+    doc: "Absolute Value (wrapping)."
+    arguments: ["a: {type[1]}"]
+    return_type: "{type[1]}"
+    attr:
+      - *neon-stable
+    assert_instr: [abs]
+    safety: safe
+    types:
+      - [i64, i64]
+    compose:
+      # FIXME(llvm): this should be `i64::wrapping_abs` but it optimizes differently.
+      # See https://github.com/llvm/llvm-project/issues/148388.
+      - LLVMLink:
+          name: "vabsd_s64"
+          links:
+             - link: "llvm.aarch64.neon.abs.i64"
+               arch: aarch64,arm64ec
+
   - name: "{type[0]}"
     doc: "Absolute Value (wrapping)."
     arguments: ["a: {type[1]}"]
@@ -13032,15 +13051,18 @@ intrinsics:
     assert_instr: [abs]
     safety: safe
     types:
-      - ['vabsd_s64', i64, i64]
       - ['vabs_s64', int64x1_t, v1i64]
       - ['vabsq_s64', int64x2_t, v2i64]
     compose:
-      - LLVMLink:
-          name: "{type[0]}"
-          links:
-            - link: "llvm.aarch64.neon.abs.{type[2]}"
-              arch: aarch64,arm64ec
+      - Let:
+          - neg
+          - "{type[1]}"
+          - FnCall: [simd_neg, [a]]
+      - Let:
+          - mask
+          - "{type[1]}"
+          - FnCall: [simd_ge, [a, neg]]
+      - FnCall: [simd_select, [mask, a, neg]]
 
   - name: "vuqadd{neon_type[0].no}"
     doc: "Signed saturating Accumulate of Unsigned value."