Merge pull request #2118 from heiher/vpick-ev-od

loongarch: Use `intrinsics::simd` for vpick{ev,od}

Author: sayantn

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

@@ -155,22 +155,6 @@ unsafe extern "unadjusted" {
     fn __lasx_xvrepl128vei_w(a: __v8i32, b: u32) -> __v8i32;
     #[link_name = "llvm.loongarch.lasx.xvrepl128vei.d"]
     fn __lasx_xvrepl128vei_d(a: __v4i64, b: u32) -> __v4i64;
-    #[link_name = "llvm.loongarch.lasx.xvpickev.b"]
-    fn __lasx_xvpickev_b(a: __v32i8, b: __v32i8) -> __v32i8;
-    #[link_name = "llvm.loongarch.lasx.xvpickev.h"]
-    fn __lasx_xvpickev_h(a: __v16i16, b: __v16i16) -> __v16i16;
-    #[link_name = "llvm.loongarch.lasx.xvpickev.w"]
-    fn __lasx_xvpickev_w(a: __v8i32, b: __v8i32) -> __v8i32;
-    #[link_name = "llvm.loongarch.lasx.xvpickev.d"]
-    fn __lasx_xvpickev_d(a: __v4i64, b: __v4i64) -> __v4i64;
-    #[link_name = "llvm.loongarch.lasx.xvpickod.b"]
-    fn __lasx_xvpickod_b(a: __v32i8, b: __v32i8) -> __v32i8;
-    #[link_name = "llvm.loongarch.lasx.xvpickod.h"]
-    fn __lasx_xvpickod_h(a: __v16i16, b: __v16i16) -> __v16i16;
-    #[link_name = "llvm.loongarch.lasx.xvpickod.w"]
-    fn __lasx_xvpickod_w(a: __v8i32, b: __v8i32) -> __v8i32;
-    #[link_name = "llvm.loongarch.lasx.xvpickod.d"]
-    fn __lasx_xvpickod_d(a: __v4i64, b: __v4i64) -> __v4i64;
     #[link_name = "llvm.loongarch.lasx.xvilvh.b"]
     fn __lasx_xvilvh_b(a: __v32i8, b: __v32i8) -> __v32i8;
     #[link_name = "llvm.loongarch.lasx.xvilvh.h"]
@@ -1653,62 +1637,6 @@ pub fn lasx_xvrepl128vei_d<const IMM1: u32>(a: m256i) -> m256i {
     unsafe { transmute(__lasx_xvrepl128vei_d(transmute(a), IMM1)) }
 }
 
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvpickev_b(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvpickev_b(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvpickev_h(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvpickev_h(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvpickev_w(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvpickev_w(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvpickev_d(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvpickev_d(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvpickod_b(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvpickod_b(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvpickod_h(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvpickod_h(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvpickod_w(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvpickod_w(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lasx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lasx_xvpickod_d(a: m256i, b: m256i) -> m256i {
-    unsafe { transmute(__lasx_xvpickod_d(transmute(a), transmute(b))) }
-}
-
 #[inline]
 #[target_feature(enable = "lasx")]
 #[unstable(feature = "stdarch_loongarch", issue = "117427")]

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

@@ -5,6 +5,68 @@ use crate::core_arch::simd::{self as cs, *};
 use crate::intrinsics::simd as is;
 use crate::mem::transmute;
 
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickev_b<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(
+        b,
+        a,
+        [
+            0, 2, 4, 6, 8, 10, 12, 14, 32, 34, 36, 38, 40, 42, 44, 46,
+            16, 18, 20, 22, 24, 26, 28, 30, 48, 50, 52, 54, 56, 58, 60, 62
+        ]
+    )
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickev_d<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [0, 4, 2, 6])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickev_w<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [0, 2, 8, 10, 4, 6, 12, 14])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickev_h<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [0, 2, 4, 6, 16, 18, 20, 22, 8, 10, 12, 14, 24, 26, 28, 30])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickod_b<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(
+        b,
+        a,
+        [
+            1, 3, 5, 7, 9, 11, 13, 15, 33, 35, 37, 39, 41, 43, 45, 47,
+            17, 19, 21, 23, 25, 27, 29, 31, 49, 51, 53, 55, 57, 59, 61, 63
+        ]
+    )
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickod_d<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [1, 5, 3, 7])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickod_w<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [1, 3, 9, 11, 5, 7, 13, 15])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickod_h<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [1, 3, 5, 7, 17, 19, 21, 23, 9, 11, 13, 15, 25, 27, 29, 31])
+}
+
 impl_vv!("lasx", lasx_xvpcnt_b, is::simd_ctpop, m256i, i8x32);
 impl_vv!("lasx", lasx_xvpcnt_h, is::simd_ctpop, m256i, i16x16);
 impl_vv!("lasx", lasx_xvpcnt_w, is::simd_ctpop, m256i, i32x8);
@@ -160,6 +222,14 @@ 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_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);
+impl_vvv!("lasx", lasx_xvpickev_d, simd_pickev_d, m256i, i64x4);
+impl_vvv!("lasx", lasx_xvpickod_b, simd_pickod_b, m256i, i8x32);
+impl_vvv!("lasx", lasx_xvpickod_h, simd_pickod_h, m256i, i16x16);
+impl_vvv!("lasx", lasx_xvpickod_w, simd_pickod_w, m256i, i32x8);
+impl_vvv!("lasx", lasx_xvpickod_d, simd_pickod_d, m256i, i64x4);
 
 impl_vuv!("lasx", lasx_xvslli_b, is::simd_shl, m256i, i8x32);
 impl_vuv!("lasx", lasx_xvslli_h, is::simd_shl, m256i, i16x16);

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

@@ -163,22 +163,6 @@ unsafe extern "unadjusted" {
     fn __lsx_vreplvei_w(a: __v4i32, b: u32) -> __v4i32;
     #[link_name = "llvm.loongarch.lsx.vreplvei.d"]
     fn __lsx_vreplvei_d(a: __v2i64, b: u32) -> __v2i64;
-    #[link_name = "llvm.loongarch.lsx.vpickev.b"]
-    fn __lsx_vpickev_b(a: __v16i8, b: __v16i8) -> __v16i8;
-    #[link_name = "llvm.loongarch.lsx.vpickev.h"]
-    fn __lsx_vpickev_h(a: __v8i16, b: __v8i16) -> __v8i16;
-    #[link_name = "llvm.loongarch.lsx.vpickev.w"]
-    fn __lsx_vpickev_w(a: __v4i32, b: __v4i32) -> __v4i32;
-    #[link_name = "llvm.loongarch.lsx.vpickev.d"]
-    fn __lsx_vpickev_d(a: __v2i64, b: __v2i64) -> __v2i64;
-    #[link_name = "llvm.loongarch.lsx.vpickod.b"]
-    fn __lsx_vpickod_b(a: __v16i8, b: __v16i8) -> __v16i8;
-    #[link_name = "llvm.loongarch.lsx.vpickod.h"]
-    fn __lsx_vpickod_h(a: __v8i16, b: __v8i16) -> __v8i16;
-    #[link_name = "llvm.loongarch.lsx.vpickod.w"]
-    fn __lsx_vpickod_w(a: __v4i32, b: __v4i32) -> __v4i32;
-    #[link_name = "llvm.loongarch.lsx.vpickod.d"]
-    fn __lsx_vpickod_d(a: __v2i64, b: __v2i64) -> __v2i64;
     #[link_name = "llvm.loongarch.lsx.vilvh.b"]
     fn __lsx_vilvh_b(a: __v16i8, b: __v16i8) -> __v16i8;
     #[link_name = "llvm.loongarch.lsx.vilvh.h"]
@@ -1593,62 +1577,6 @@ pub fn lsx_vreplvei_d<const IMM1: u32>(a: m128i) -> m128i {
     unsafe { transmute(__lsx_vreplvei_d(transmute(a), IMM1)) }
 }
 
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vpickev_b(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vpickev_b(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vpickev_h(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vpickev_h(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vpickev_w(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vpickev_w(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vpickev_d(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vpickev_d(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vpickod_b(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vpickod_b(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vpickod_h(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vpickod_h(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vpickod_w(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vpickod_w(transmute(a), transmute(b))) }
-}
-
-#[inline]
-#[target_feature(enable = "lsx")]
-#[unstable(feature = "stdarch_loongarch", issue = "117427")]
-pub fn lsx_vpickod_d(a: m128i, b: m128i) -> m128i {
-    unsafe { transmute(__lsx_vpickod_d(transmute(a), transmute(b))) }
-}
-
 #[inline]
 #[target_feature(enable = "lsx")]
 #[unstable(feature = "stdarch_loongarch", issue = "117427")]

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

@@ -5,6 +5,54 @@ use crate::core_arch::simd::{self as cs, *};
 use crate::intrinsics::simd as is;
 use crate::mem::transmute;
 
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickev_b<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickev_h<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [0, 2, 4, 6, 8, 10, 12, 14])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickev_w<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [0, 2, 4, 6])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickev_d<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [0, 2])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickod_b<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickod_h<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [1, 3, 5, 7, 9, 11, 13, 15])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickod_w<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [1, 3, 5, 7])
+}
+
+#[inline(always)]
+#[rustc_const_unstable(feature = "stdarch_const_helpers", issue = "none")]
+const unsafe fn simd_pickod_d<T: Copy>(a: T, b: T) -> T {
+    simd_shuffle!(b, a, [1, 3])
+}
+
 impl_vv!("lsx", lsx_vpcnt_b, is::simd_ctpop, m128i, i8x16);
 impl_vv!("lsx", lsx_vpcnt_h, is::simd_ctpop, m128i, i16x8);
 impl_vv!("lsx", lsx_vpcnt_w, is::simd_ctpop, m128i, i32x4);
@@ -160,6 +208,14 @@ 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_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);
+impl_vvv!("lsx", lsx_vpickev_d, simd_pickev_d, m128i, i64x2);
+impl_vvv!("lsx", lsx_vpickod_b, simd_pickod_b, m128i, i8x16);
+impl_vvv!("lsx", lsx_vpickod_h, simd_pickod_h, m128i, i16x8);
+impl_vvv!("lsx", lsx_vpickod_w, simd_pickod_w, m128i, i32x4);
+impl_vvv!("lsx", lsx_vpickod_d, simd_pickod_d, m128i, i64x2);
 
 impl_vuv!("lsx", lsx_vslli_b, is::simd_shl, m128i, i8x16);
 impl_vuv!("lsx", lsx_vslli_h, is::simd_shl, m128i, i16x8);

crates/stdarch-gen-loongarch/lasx.spec

@@ -1420,41 +1420,49 @@ asm-fmts = xd, xj, ui1
 data-types = V4DI, V4DI, UQI
 
 /// lasx_xvpickev_b
+impl = portable
 name = lasx_xvpickev_b
 asm-fmts = xd, xj, xk
 data-types = V32QI, V32QI, V32QI
 
 /// lasx_xvpickev_h
+impl = portable
 name = lasx_xvpickev_h
 asm-fmts = xd, xj, xk
 data-types = V16HI, V16HI, V16HI
 
 /// lasx_xvpickev_w
+impl = portable
 name = lasx_xvpickev_w
 asm-fmts = xd, xj, xk
 data-types = V8SI, V8SI, V8SI
 
 /// lasx_xvpickev_d
+impl = portable
 name = lasx_xvpickev_d
 asm-fmts = xd, xj, xk
 data-types = V4DI, V4DI, V4DI
 
 /// lasx_xvpickod_b
+impl = portable
 name = lasx_xvpickod_b
 asm-fmts = xd, xj, xk
 data-types = V32QI, V32QI, V32QI
 
 /// lasx_xvpickod_h
+impl = portable
 name = lasx_xvpickod_h
 asm-fmts = xd, xj, xk
 data-types = V16HI, V16HI, V16HI
 
 /// lasx_xvpickod_w
+impl = portable
 name = lasx_xvpickod_w
 asm-fmts = xd, xj, xk
 data-types = V8SI, V8SI, V8SI
 
 /// lasx_xvpickod_d
+impl = portable
 name = lasx_xvpickod_d
 asm-fmts = xd, xj, xk
 data-types = V4DI, V4DI, V4DI