loongarch: Use intrinsics::simd for vfrsqrt

crates/core_arch/src/loongarch64/lasx/generated.rs

@@ -211,10 +211,6 @@ unsafe extern "unadjusted" {
     fn __lasx_xvfrint_s(a: __v8f32) -> __v8f32;
     #[link_name = "llvm.loongarch.lasx.xvfrint.d"]
     fn __lasx_xvfrint_d(a: __v4f64) -> __v4f64;
-    #[link_name = "llvm.loongarch.lasx.xvfrsqrt.s"]
-    fn __lasx_xvfrsqrt_s(a: __v8f32) -> __v8f32;
-    #[link_name = "llvm.loongarch.lasx.xvfrsqrt.d"]
-    fn __lasx_xvfrsqrt_d(a: __v4f64) -> __v4f64;
     #[link_name = "llvm.loongarch.lasx.xvflogb.s"]
     fn __lasx_xvflogb_s(a: __v8f32) -> __v8f32;
     #[link_name = "llvm.loongarch.lasx.xvflogb.d"]
@@ -1751,20 +1747,6 @@ pub fn lasx_xvfrint_d(a: m256d) -> m256d {
     unsafe { transmute(__lasx_xvfrint_d(transmute(a))) }
 }
 
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvfrsqrt_s(a: m256) -> m256 {
-    unsafe { transmute(__lasx_xvfrsqrt_s(transmute(a))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvfrsqrt_d(a: m256d) -> m256d {
-    unsafe { transmute(__lasx_xvfrsqrt_d(transmute(a))) }
-}
-
 #[inline]
 #[target_feature(enable = "lasx")]
 #[unstable(feature = "stdarch_loongarch", issue = "117427")]

crates/core_arch/src/loongarch64/lasx/portable.rs

@@ -328,6 +328,8 @@ impl_vv!("lasx", lasx_xvneg_w, is::simd_neg, m256i, i32x8);
 impl_vv!("lasx", lasx_xvneg_d, is::simd_neg, m256i, i64x4);
 impl_vv!("lasx", lasx_xvfsqrt_s, is::simd_fsqrt, m256, f32x8);
 impl_vv!("lasx", lasx_xvfsqrt_d, is::simd_fsqrt, m256d, f64x4);
+impl_vv!("lasx", lasx_xvfrsqrt_s, ls::simd_frsqrt_s, m256, f32x8);
+impl_vv!("lasx", lasx_xvfrsqrt_d, ls::simd_frsqrt_d, m256d, f64x4);
 impl_vv!("lasx", lasx_xvfrecip_s, ls::simd_frecip_s, m256, f32x8);
 impl_vv!("lasx", lasx_xvfrecip_d, ls::simd_frecip_d, m256d, f64x4);
 impl_vv!("lasx", lasx_xvreplve0_b, simd_replve0_b, m256i, i8x32);

crates/core_arch/src/loongarch64/lsx/generated.rs

@@ -217,10 +217,6 @@ unsafe extern "unadjusted" {
     fn __lsx_vfrint_s(a: __v4f32) -> __v4f32;
     #[link_name = "llvm.loongarch.lsx.vfrint.d"]
     fn __lsx_vfrint_d(a: __v2f64) -> __v2f64;
-    #[link_name = "llvm.loongarch.lsx.vfrsqrt.s"]
-    fn __lsx_vfrsqrt_s(a: __v4f32) -> __v4f32;
-    #[link_name = "llvm.loongarch.lsx.vfrsqrt.d"]
-    fn __lsx_vfrsqrt_d(a: __v2f64) -> __v2f64;
     #[link_name = "llvm.loongarch.lsx.vflogb.s"]
     fn __lsx_vflogb_s(a: __v4f32) -> __v4f32;
     #[link_name = "llvm.loongarch.lsx.vflogb.d"]
@@ -1692,20 +1688,6 @@ pub fn lsx_vfrint_d(a: m128d) -> m128d {
     unsafe { transmute(__lsx_vfrint_d(transmute(a))) }
 }
 
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vfrsqrt_s(a: m128) -> m128 {
-    unsafe { transmute(__lsx_vfrsqrt_s(transmute(a))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vfrsqrt_d(a: m128d) -> m128d {
-    unsafe { transmute(__lsx_vfrsqrt_d(transmute(a))) }
-}
-
 #[inline]
 #[target_feature(enable = "lsx")]
 #[unstable(feature = "stdarch_loongarch", issue = "117427")]

crates/core_arch/src/loongarch64/lsx/portable.rs

@@ -227,6 +227,8 @@ impl_vv!("lsx", lsx_vneg_w, is::simd_neg, m128i, i32x4);
 impl_vv!("lsx", lsx_vneg_d, is::simd_neg, m128i, i64x2);
 impl_vv!("lsx", lsx_vfsqrt_s, is::simd_fsqrt, m128, f32x4);
 impl_vv!("lsx", lsx_vfsqrt_d, is::simd_fsqrt, m128d, f64x2);
+impl_vv!("lsx", lsx_vfrsqrt_s, ls::simd_frsqrt_s, m128, f32x4);
+impl_vv!("lsx", lsx_vfrsqrt_d, ls::simd_frsqrt_d, m128d, f64x2);
 impl_vv!("lsx", lsx_vfrecip_s, ls::simd_frecip_s, m128, f32x4);
 impl_vv!("lsx", lsx_vfrecip_d, ls::simd_frecip_d, m128d, f64x2);
 

crates/core_arch/src/loongarch64/simd.rs

@@ -120,6 +120,16 @@ pub(super) const unsafe fn simd_frecip_d<T: Copy>(a: T) -> T {
     is::simd_div(is::simd_splat(1.0f64), a)
 }
 
+#[inline(always)]
+pub(super) unsafe fn simd_frsqrt_s<T: Copy>(a: T) -> T {
+    ls::simd_frecip_s(is::simd_fsqrt(a))
+}
+
+#[inline(always)]
+pub(super) unsafe fn simd_frsqrt_d<T: Copy>(a: T) -> T {
+    ls::simd_frecip_d(is::simd_fsqrt(a))
+}
+
 #[inline(always)]
 #[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
 pub(super) const unsafe fn simd_ld<const I: i32, T: Copy>(a: *const i8) -> T {

crates/stdarch-gen-loongarch/lasx.spec

@@ -1914,11 +1914,13 @@ asm-fmts = xd, xj
 data-types = V4DF, V4DF
 
 /// lasx_xvfrsqrt_s
+impl = portable
 name = lasx_xvfrsqrt_s
 asm-fmts = xd, xj
 data-types = V8SF, V8SF
 
 /// lasx_xvfrsqrt_d
+impl = portable
 name = lasx_xvfrsqrt_d
 asm-fmts = xd, xj
 data-types = V4DF, V4DF

crates/stdarch-gen-loongarch/lsx.spec

@@ -2001,11 +2001,13 @@ asm-fmts = vd, vj
 data-types = V2DF, V2DF
 
 /// lsx_vfrsqrt_s
+impl = portable
 name = lsx_vfrsqrt_s
 asm-fmts = vd, vj
 data-types = V4SF, V4SF
 
 /// lsx_vfrsqrt_d
+impl = portable
 name = lsx_vfrsqrt_d
 asm-fmts = vd, vj
 data-types = V2DF, V2DF

crates/stdarch-gen-loongarch/src/portable-intrinsics.txt

@@ -277,6 +277,8 @@ lsx_vrotri_w
 lsx_vrotri_d
 lsx_vfrecip_s
 lsx_vfrecip_d
+lsx_vfrsqrt_s
+lsx_vfrsqrt_d
 
 # LASX intrinsics
 lasx_xvsll_b
@@ -555,3 +557,5 @@ lasx_xvrotri_w
 lasx_xvrotri_d
 lasx_xvfrecip_s
 lasx_xvfrecip_d
+lasx_xvfrsqrt_s
+lasx_xvfrsqrt_d