loongarch: Use intrinsics::simd for vmuh

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

@@ -257,22 +257,6 @@ unsafe extern "unadjusted" {
     fn __lasx_xvreplve_d(a: __v4i64, b: i32) -> __v4i64;
     #[link_name = "llvm.loongarch.lasx.xvpermi.w"]
     fn __lasx_xvpermi_w(a: __v8i32, b: __v8i32, c: u32) -> __v8i32;
-    #[link_name = "llvm.loongarch.lasx.xvmuh.b"]
-    fn __lasx_xvmuh_b(a: __v32i8, b: __v32i8) -> __v32i8;
-    #[link_name = "llvm.loongarch.lasx.xvmuh.h"]
-    fn __lasx_xvmuh_h(a: __v16i16, b: __v16i16) -> __v16i16;
-    #[link_name = "llvm.loongarch.lasx.xvmuh.w"]
-    fn __lasx_xvmuh_w(a: __v8i32, b: __v8i32) -> __v8i32;
-    #[link_name = "llvm.loongarch.lasx.xvmuh.d"]
-    fn __lasx_xvmuh_d(a: __v4i64, b: __v4i64) -> __v4i64;
-    #[link_name = "llvm.loongarch.lasx.xvmuh.bu"]
-    fn __lasx_xvmuh_bu(a: __v32u8, b: __v32u8) -> __v32u8;
-    #[link_name = "llvm.loongarch.lasx.xvmuh.hu"]
-    fn __lasx_xvmuh_hu(a: __v16u16, b: __v16u16) -> __v16u16;
-    #[link_name = "llvm.loongarch.lasx.xvmuh.wu"]
-    fn __lasx_xvmuh_wu(a: __v8u32, b: __v8u32) -> __v8u32;
-    #[link_name = "llvm.loongarch.lasx.xvmuh.du"]
-    fn __lasx_xvmuh_du(a: __v4u64, b: __v4u64) -> __v4u64;
     #[link_name = "llvm.loongarch.lasx.xvsllwil.h.b"]
     fn __lasx_xvsllwil_h_b(a: __v32i8, b: u32) -> __v16i16;
     #[link_name = "llvm.loongarch.lasx.xvsllwil.w.h"]
@@ -1898,62 +1882,6 @@ pub fn lasx_xvpermi_w<const IMM8: u32>(a: m256i, b: m256i) -> m256i {
     unsafe { transmute(__lasx_xvpermi_w(transmute(a), transmute(b), IMM8)) }
 }
 
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvmuh_b(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvmuh_b(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvmuh_h(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvmuh_h(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvmuh_w(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvmuh_w(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvmuh_d(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvmuh_d(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvmuh_bu(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvmuh_bu(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvmuh_hu(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvmuh_hu(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvmuh_wu(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvmuh_wu(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvmuh_du(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvmuh_du(transmute(a), transmute(b))) }
-}
-
 #[inline]
 #[target_feature(enable = "lasx")]
 #[rustc_legacy_const_generics(1)]

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

@@ -492,6 +492,14 @@ impl_vvv!("lasx", lasx_xvabsd_bu, ls::simd_absd, m256i, u8x32);
 impl_vvv!("lasx", lasx_xvabsd_hu, ls::simd_absd, m256i, u16x16);
 impl_vvv!("lasx", lasx_xvabsd_wu, ls::simd_absd, m256i, u32x8);
 impl_vvv!("lasx", lasx_xvabsd_du, ls::simd_absd, m256i, u64x4);
+impl_vvv!("lasx", lasx_xvmuh_b, simd_muh, m256i, i8x32, i16x32);
+impl_vvv!("lasx", lasx_xvmuh_h, simd_muh, m256i, i16x16, i32x16);
+impl_vvv!("lasx", lasx_xvmuh_w, simd_muh, m256i, i32x8, i64x8);
+impl_vvv!("lasx", lasx_xvmuh_d, simd_muh, m256i, i64x4, i128x4);
+impl_vvv!("lasx", lasx_xvmuh_bu, simd_muh, m256i, u8x32, u16x32);
+impl_vvv!("lasx", lasx_xvmuh_hu, simd_muh, m256i, u16x16, u32x16);
+impl_vvv!("lasx", lasx_xvmuh_wu, simd_muh, m256i, u32x8, u64x8);
+impl_vvv!("lasx", lasx_xvmuh_du, simd_muh, m256i, u64x4, u128x4);
 impl_vvv!("lasx", lasx_xvpickev_b, simd_pickev_b, m256i, i8x32);
 impl_vvv!("lasx", lasx_xvpickev_h, simd_pickev_h, m256i, i16x16);
 impl_vvv!("lasx", lasx_xvpickev_w, simd_pickev_w, m256i, i32x8);

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

@@ -253,22 +253,6 @@ unsafe extern "unadjusted" {
     fn __lsx_vffint_s_wu(a: __v4u32) -> __v4f32;
     #[link_name = "llvm.loongarch.lsx.vffint.d.lu"]
     fn __lsx_vffint_d_lu(a: __v2u64) -> __v2f64;
-    #[link_name = "llvm.loongarch.lsx.vmuh.b"]
-    fn __lsx_vmuh_b(a: __v16i8, b: __v16i8) -> __v16i8;
-    #[link_name = "llvm.loongarch.lsx.vmuh.h"]
-    fn __lsx_vmuh_h(a: __v8i16, b: __v8i16) -> __v8i16;
-    #[link_name = "llvm.loongarch.lsx.vmuh.w"]
-    fn __lsx_vmuh_w(a: __v4i32, b: __v4i32) -> __v4i32;
-    #[link_name = "llvm.loongarch.lsx.vmuh.d"]
-    fn __lsx_vmuh_d(a: __v2i64, b: __v2i64) -> __v2i64;
-    #[link_name = "llvm.loongarch.lsx.vmuh.bu"]
-    fn __lsx_vmuh_bu(a: __v16u8, b: __v16u8) -> __v16u8;
-    #[link_name = "llvm.loongarch.lsx.vmuh.hu"]
-    fn __lsx_vmuh_hu(a: __v8u16, b: __v8u16) -> __v8u16;
-    #[link_name = "llvm.loongarch.lsx.vmuh.wu"]
-    fn __lsx_vmuh_wu(a: __v4u32, b: __v4u32) -> __v4u32;
-    #[link_name = "llvm.loongarch.lsx.vmuh.du"]
-    fn __lsx_vmuh_du(a: __v2u64, b: __v2u64) -> __v2u64;
     #[link_name = "llvm.loongarch.lsx.vsllwil.h.b"]
     fn __lsx_vsllwil_h_b(a: __v16i8, b: u32) -> __v8i16;
     #[link_name = "llvm.loongarch.lsx.vsllwil.w.h"]
@@ -1802,62 +1786,6 @@ pub fn lsx_vffint_d_lu(a: m128i) -> m128d {
     unsafe { transmute(__lsx_vffint_d_lu(transmute(a))) }
 }
 
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vmuh_b(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vmuh_b(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vmuh_h(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vmuh_h(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vmuh_w(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vmuh_w(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vmuh_d(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vmuh_d(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vmuh_bu(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vmuh_bu(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vmuh_hu(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vmuh_hu(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vmuh_wu(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vmuh_wu(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vmuh_du(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vmuh_du(transmute(a), transmute(b))) }
-}
-
 #[inline]
 #[target_feature(enable = "lsx")]
 #[rustc_legacy_const_generics(1)]

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

@@ -386,6 +386,14 @@ impl_vvv!("lsx", lsx_vabsd_bu, ls::simd_absd, m128i, u8x16);
 impl_vvv!("lsx", lsx_vabsd_hu, ls::simd_absd, m128i, u16x8);
 impl_vvv!("lsx", lsx_vabsd_wu, ls::simd_absd, m128i, u32x4);
 impl_vvv!("lsx", lsx_vabsd_du, ls::simd_absd, m128i, u64x2);
+impl_vvv!("lsx", lsx_vmuh_b, simd_muh, m128i, i8x16, i16x16);
+impl_vvv!("lsx", lsx_vmuh_h, simd_muh, m128i, i16x8, i32x8);
+impl_vvv!("lsx", lsx_vmuh_w, simd_muh, m128i, i32x4, i64x4);
+impl_vvv!("lsx", lsx_vmuh_d, simd_muh, m128i, i64x2, i128x2);
+impl_vvv!("lsx", lsx_vmuh_bu, simd_muh, m128i, u8x16, u16x16);
+impl_vvv!("lsx", lsx_vmuh_hu, simd_muh, m128i, u16x8, u32x8);
+impl_vvv!("lsx", lsx_vmuh_wu, simd_muh, m128i, u32x4, u64x4);
+impl_vvv!("lsx", lsx_vmuh_du, simd_muh, m128i, u64x2, u128x2);
 impl_vvv!("lsx", lsx_vpickev_b, simd_pickev_b, m128i, i8x16);
 impl_vvv!("lsx", lsx_vpickev_h, simd_pickev_h, m128i, i16x8);
 impl_vvv!("lsx", lsx_vpickev_w, simd_pickev_w, m128i, i32x4);

crates/core_arch/src/loongarch64/simd.rs

@@ -35,16 +35,26 @@ impl_simd_ext!(u8x16, u8);
 impl_simd_ext!(u8x32, u8);
 impl_simd_ext!(i16x8, i16);
 impl_simd_ext!(i16x16, i16);
+impl_simd_ext!(i16x32, i16);
 impl_simd_ext!(u16x8, u16);
 impl_simd_ext!(u16x16, u16);
+impl_simd_ext!(u16x32, u16);
 impl_simd_ext!(i32x4, i32);
 impl_simd_ext!(i32x8, i32);
+impl_simd_ext!(i32x16, i32);
 impl_simd_ext!(u32x4, u32);
 impl_simd_ext!(u32x8, u32);
+impl_simd_ext!(u32x16, u32);
 impl_simd_ext!(i64x2, i64);
 impl_simd_ext!(i64x4, i64);
+impl_simd_ext!(i64x8, i64);
 impl_simd_ext!(u64x2, u64);
 impl_simd_ext!(u64x4, u64);
+impl_simd_ext!(u64x8, u64);
+impl_simd_ext!(i128x2, i128);
+impl_simd_ext!(u128x2, u128);
+impl_simd_ext!(i128x4, i128);
+impl_simd_ext!(u128x4, u128);
 
 #[inline(always)]
 #[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
@@ -156,6 +166,18 @@ pub(super) const unsafe fn simd_msub<T: Copy>(a: T, b: T, c: T) -> T {
     is::simd_sub(a, is::simd_mul(b, c))
 }
 
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+pub(super) const unsafe fn simd_muh<T: Copy, W: Copy + const SimdExt>(a: T, b: T) -> T {
+    let a: W = is::simd_cast(a);
+    let b: W = is::simd_cast(b);
+    let p = is::simd_mul(a, b);
+    is::simd_cast(is::simd_shr(
+        p,
+        ls::simd_splat((size_of::<W::Elem>() * 8 / 2) as i64),
+    ))
+}
+
 #[inline(always)]
 #[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
 pub(super) const unsafe fn simd_nor<T: Copy + const SimdExt>(a: T, b: T) -> T {
@@ -312,6 +334,19 @@ macro_rules! impl_vvv {
             }
         }
     };
+    ($ft:literal, $name:ident, $op:ident, $oty:ty, $ity:ty, $wty:ty) => {
+        #[inline]
+        #[target_feature(enable = $ft)]
+        #[unstable(feature = "stdarch_loongarch", issue = "117427")]
+        pub fn $name(a: $oty, b: $oty) -> $oty {
+            unsafe {
+                let a: $ity = transmute(a);
+                let b: $ity = transmute(b);
+                let r: $ity = $op::<$ity, $wty>(a, b);
+                transmute(r)
+            }
+        }
+    };
 }
 
 pub(super) use impl_vvv;

crates/core_arch/src/simd.rs

@@ -28,11 +28,13 @@ unsafe impl SimdElement for u8 {}
 unsafe impl SimdElement for u16 {}
 unsafe impl SimdElement for u32 {}
 unsafe impl SimdElement for u64 {}
+unsafe impl SimdElement for u128 {}
 
 unsafe impl SimdElement for i8 {}
 unsafe impl SimdElement for i16 {}
 unsafe impl SimdElement for i32 {}
 unsafe impl SimdElement for i64 {}
+unsafe impl SimdElement for i128 {}
 
 unsafe impl SimdElement for f16 {}
 unsafe impl SimdElement for f32 {}
@@ -371,11 +373,13 @@ pub(crate) type u8x32 = Simd<u8, 32>;
 pub(crate) type u16x16 = Simd<u16, 16>;
 pub(crate) type u32x8 = Simd<u32, 8>;
 pub(crate) type u64x4 = Simd<u64, 4>;
+pub(crate) type u128x2 = Simd<u128, 2>;
 
 pub(crate) type i8x32 = Simd<i8, 32>;
 pub(crate) type i16x16 = Simd<i16, 16>;
 pub(crate) type i32x8 = Simd<i32, 8>;
 pub(crate) type i64x4 = Simd<i64, 4>;
+pub(crate) type i128x2 = Simd<i128, 2>;
 
 pub(crate) type f16x16 = Simd<f16, 16>;
 pub(crate) type f32x8 = Simd<f32, 8>;
@@ -391,11 +395,13 @@ pub(crate) type u8x64 = Simd<u8, 64>;
 pub(crate) type u16x32 = Simd<u16, 32>;
 pub(crate) type u32x16 = Simd<u32, 16>;
 pub(crate) type u64x8 = Simd<u64, 8>;
+pub(crate) type u128x4 = Simd<u128, 4>;
 
 pub(crate) type i8x64 = Simd<i8, 64>;
 pub(crate) type i16x32 = Simd<i16, 32>;
 pub(crate) type i32x16 = Simd<i32, 16>;
 pub(crate) type i64x8 = Simd<i64, 8>;
+pub(crate) type i128x4 = Simd<i128, 4>;
 
 pub(crate) type f16x32 = Simd<f16, 32>;
 pub(crate) type f32x16 = Simd<f32, 16>;

crates/stdarch-gen-loongarch/lasx.spec

@@ -2071,41 +2071,49 @@ asm-fmts = xd, xj
 data-types = V4DI, V4DI
 
 /// lasx_xvmuh_b
+impl = portable
 name = lasx_xvmuh_b
 asm-fmts = xd, xj, xk
 data-types = V32QI, V32QI, V32QI
 
 /// lasx_xvmuh_h
+impl = portable
 name = lasx_xvmuh_h
 asm-fmts = xd, xj, xk
 data-types = V16HI, V16HI, V16HI
 
 /// lasx_xvmuh_w
+impl = portable
 name = lasx_xvmuh_w
 asm-fmts = xd, xj, xk
 data-types = V8SI, V8SI, V8SI
 
 /// lasx_xvmuh_d
+impl = portable
 name = lasx_xvmuh_d
 asm-fmts = xd, xj, xk
 data-types = V4DI, V4DI, V4DI
 
 /// lasx_xvmuh_bu
+impl = portable
 name = lasx_xvmuh_bu
 asm-fmts = xd, xj, xk
 data-types = UV32QI, UV32QI, UV32QI
 
 /// lasx_xvmuh_hu
+impl = portable
 name = lasx_xvmuh_hu
 asm-fmts = xd, xj, xk
 data-types = UV16HI, UV16HI, UV16HI
 
 /// lasx_xvmuh_wu
+impl = portable
 name = lasx_xvmuh_wu
 asm-fmts = xd, xj, xk
 data-types = UV8SI, UV8SI, UV8SI
 
 /// lasx_xvmuh_du
+impl = portable
 name = lasx_xvmuh_du
 asm-fmts = xd, xj, xk
 data-types = UV4DI, UV4DI, UV4DI