Add __extendhfdf2

LLVM doesn't seem to emit this intrinsic but it probably should, in some
cases it lowers f16->f64 conversions as f16->f32->f64 with two libcalls.
GCC provides this intrinsic so it is good to have anyway.

[1]: https://rust.godbolt.org/z/xezM9PEnz

Author: tgross35

src/float/extend.rs

@@ -96,6 +96,14 @@ intrinsics! {
         extend(a)
     }
 
+    #[avr_skip]
+    #[aapcs_on_arm]
+    #[apple_f16_arg_abi]
+    #[cfg(f16_enabled)]
+    pub extern "C" fn __extendhfdf2(a: f16) -> f64 {
+        extend(a)
+    }
+
     #[avr_skip]
     #[aapcs_on_arm]
     #[ppc_alias = __extendhfkf2]

testcrate/benches/float_extend.rs

@@ -28,6 +28,28 @@ float_bench! {
     ],
 }
 
+#[cfg(f16_enabled)]
+float_bench! {
+    name: extend_f16_f64,
+    sig: (a: f16) -> f64,
+    crate_fn: extend::__extendhfdf2,
+    sys_fn: __extendhfdf2,
+    sys_available: not(feature = "no-sys-f16"),
+    asm: [
+        #[cfg(target_arch = "aarch64")] {
+            let ret: f64;
+            asm!(
+                "fcvt    {ret:d}, {a:h}",
+                a = in(vreg) a,
+                ret = lateout(vreg) ret,
+                options(nomem, nostack, pure),
+            );
+
+            ret
+        };
+    ],
+}
+
 #[cfg(all(f16_enabled, f128_enabled))]
 float_bench! {
     name: extend_f16_f128,
@@ -93,6 +115,7 @@ pub fn float_extend() {
     #[cfg(not(any(target_arch = "powerpc", target_arch = "powerpc64")))]
     {
         extend_f16_f32(&mut criterion);
+        extend_f16_f64(&mut criterion);
 
         #[cfg(f128_enabled)]
         extend_f16_f128(&mut criterion);

testcrate/tests/conv.rs

@@ -311,6 +311,7 @@ mod extend {
         extend,
         f16 => f32, Half => Single, __extendhfsf2, not(feature = "no-sys-f16");
         f16 => f32, Half => Single, __gnu_h2f_ieee, not(feature = "no-sys-f16");
+        f16 => f64, Half => Double, __extendhfdf2, not(feature = "no-sys-f16");
         f16 => f128, Half => Quad, __extendhftf2, not(feature = "no-sys-f16-f128-convert");
         f32 => f128, Single => Quad, __extendsftf2, not(feature = "no-sys-f128");
         f64 => f128, Double => Quad, __extenddftf2, not(feature = "no-sys-f128");