From 9835e2ae736235810b4ea1c162ca5e65c547e770 Mon Sep 17 00:00:00 2001
From: Daniel Baumann <daniel.baumann@progress-linux.org>
Date: Sat, 18 May 2024 04:49:50 +0200
Subject: Merging upstream version 1.71.1+dfsg1.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
---
 vendor/ppv-lite86/src/generic.rs     | 158 ++++++++-----
 vendor/ppv-lite86/src/lib.rs         |   8 +-
 vendor/ppv-lite86/src/soft.rs        |  92 ++++++--
 vendor/ppv-lite86/src/types.rs       | 321 ++++++++++++++------------
 vendor/ppv-lite86/src/x86_64/mod.rs  |  32 +--
 vendor/ppv-lite86/src/x86_64/sse2.rs | 431 ++++++++++++++++++++---------------
 6 files changed, 611 insertions(+), 431 deletions(-)

(limited to 'vendor/ppv-lite86/src')

diff --git a/vendor/ppv-lite86/src/generic.rs b/vendor/ppv-lite86/src/generic.rs
index 4f4113fc3..add6c4856 100644
--- a/vendor/ppv-lite86/src/generic.rs
+++ b/vendor/ppv-lite86/src/generic.rs
@@ -1,50 +1,50 @@
 #![allow(non_camel_case_types)]
 
-use core::ops::*;
 use crate::soft::{x2, x4};
 use crate::types::*;
+use core::ops::*;
 
+#[repr(C)]
 #[derive(Clone, Copy)]
 pub union vec128_storage {
     d: [u32; 4],
     q: [u64; 2],
-    o: [u128; 1],
 }
 impl From<[u32; 4]> for vec128_storage {
-    #[inline]
+    #[inline(always)]
     fn from(d: [u32; 4]) -> Self {
         Self { d }
     }
 }
 impl From<vec128_storage> for [u32; 4] {
-    #[inline]
+    #[inline(always)]
     fn from(d: vec128_storage) -> Self {
         unsafe { d.d }
     }
 }
 impl From<[u64; 2]> for vec128_storage {
-    #[inline]
+    #[inline(always)]
     fn from(q: [u64; 2]) -> Self {
         Self { q }
     }
 }
 impl From<vec128_storage> for [u64; 2] {
-    #[inline]
+    #[inline(always)]
     fn from(q: vec128_storage) -> Self {
         unsafe { q.q }
     }
 }
 impl Default for vec128_storage {
-    #[inline]
+    #[inline(always)]
     fn default() -> Self {
-        Self { o: [0] }
+        Self { q: [0, 0] }
     }
 }
 impl Eq for vec128_storage {}
 impl PartialEq<vec128_storage> for vec128_storage {
-    #[inline]
+    #[inline(always)]
     fn eq(&self, rhs: &Self) -> bool {
-        unsafe { self.o == rhs.o }
+        unsafe { self.q == rhs.q }
     }
 }
 #[derive(Clone, Copy, PartialEq, Eq, Default)]
@@ -61,20 +61,22 @@ impl vec256_storage {
         self.v128
     }
 }
-impl From<[u64; 4]> for vec256_storage {
-    #[inline]
-    fn from(q: [u64; 4]) -> Self {
-        Self { v128: [[0, 1].into(), [2, 3].into()] }
-    }
-}
 impl From<vec256_storage> for [u64; 4] {
-    #[inline]
+    #[inline(always)]
     fn from(q: vec256_storage) -> Self {
         let [a, b]: [u64; 2] = q.v128[0].into();
         let [c, d]: [u64; 2] = q.v128[1].into();
         [a, b, c, d]
     }
 }
+impl From<[u64; 4]> for vec256_storage {
+    #[inline(always)]
+    fn from([a, b, c, d]: [u64; 4]) -> Self {
+        Self {
+            v128: [[a, b].into(), [c, d].into()],
+        }
+    }
+}
 #[derive(Clone, Copy, PartialEq, Eq, Default)]
 pub struct vec512_storage {
     v128: [vec128_storage; 4],
@@ -90,6 +92,7 @@ impl vec512_storage {
     }
 }
 
+#[inline(always)]
 fn dmap<T, F>(t: T, f: F) -> T
 where
     T: Store<vec128_storage> + Into<vec128_storage>,
@@ -123,6 +126,7 @@ where
     unsafe { T::unpack(d) }
 }
 
+#[inline(always)]
 fn qmap<T, F>(t: T, f: F) -> T
 where
     T: Store<vec128_storage> + Into<vec128_storage>,
@@ -136,6 +140,7 @@ where
     unsafe { T::unpack(q) }
 }
 
+#[inline(always)]
 fn qmap2<T, F>(a: T, b: T, f: F) -> T
 where
     T: Store<vec128_storage> + Into<vec128_storage>,
@@ -151,17 +156,29 @@ where
     unsafe { T::unpack(q) }
 }
 
+#[inline(always)]
+fn o_of_q(q: [u64; 2]) -> u128 {
+    u128::from(q[0]) | (u128::from(q[1]) << 64)
+}
+
+#[inline(always)]
+fn q_of_o(o: u128) -> [u64; 2] {
+    [o as u64, (o >> 64) as u64]
+}
+
+#[inline(always)]
 fn omap<T, F>(a: T, f: F) -> T
 where
     T: Store<vec128_storage> + Into<vec128_storage>,
     F: Fn(u128) -> u128,
 {
     let a: vec128_storage = a.into();
-    let ao = unsafe { a.o };
-    let o = vec128_storage { o: [f(ao[0])] };
+    let ao = o_of_q(unsafe { a.q });
+    let o = vec128_storage { q: q_of_o(f(ao)) };
     unsafe { T::unpack(o) }
 }
 
+#[inline(always)]
 fn omap2<T, F>(a: T, b: T, f: F) -> T
 where
     T: Store<vec128_storage> + Into<vec128_storage>,
@@ -169,10 +186,10 @@ where
 {
     let a: vec128_storage = a.into();
     let b: vec128_storage = b.into();
-    let ao = unsafe { a.o };
-    let bo = unsafe { b.o };
+    let ao = o_of_q(unsafe { a.q });
+    let bo = o_of_q(unsafe { b.q });
     let o = vec128_storage {
-        o: [f(ao[0], bo[0])],
+        q: q_of_o(f(ao, bo)),
     };
     unsafe { T::unpack(o) }
 }
@@ -245,39 +262,39 @@ macro_rules! impl_bitops {
         }
 
         impl Swap64 for $vec {
-            #[inline]
+            #[inline(always)]
             fn swap1(self) -> Self {
                 qmap(self, |x| {
                     ((x & 0x5555555555555555) << 1) | ((x & 0xaaaaaaaaaaaaaaaa) >> 1)
                 })
             }
-            #[inline]
+            #[inline(always)]
             fn swap2(self) -> Self {
                 qmap(self, |x| {
                     ((x & 0x3333333333333333) << 2) | ((x & 0xcccccccccccccccc) >> 2)
                 })
             }
-            #[inline]
+            #[inline(always)]
             fn swap4(self) -> Self {
                 qmap(self, |x| {
                     ((x & 0x0f0f0f0f0f0f0f0f) << 4) | ((x & 0xf0f0f0f0f0f0f0f0) >> 4)
                 })
             }
-            #[inline]
+            #[inline(always)]
             fn swap8(self) -> Self {
                 qmap(self, |x| {
                     ((x & 0x00ff00ff00ff00ff) << 8) | ((x & 0xff00ff00ff00ff00) >> 8)
                 })
             }
-            #[inline]
+            #[inline(always)]
             fn swap16(self) -> Self {
                 dmap(self, |x| x.rotate_left(16))
             }
-            #[inline]
+            #[inline(always)]
             fn swap32(self) -> Self {
                 qmap(self, |x| x.rotate_left(32))
             }
-            #[inline]
+            #[inline(always)]
             fn swap64(self) -> Self {
                 omap(self, |x| (x << 64) | (x >> 64))
             }
@@ -289,82 +306,83 @@ impl_bitops!(u64x2_generic);
 impl_bitops!(u128x1_generic);
 
 impl RotateEachWord32 for u32x4_generic {
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right7(self) -> Self {
         dmap(self, |x| x.rotate_right(7))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right8(self) -> Self {
         dmap(self, |x| x.rotate_right(8))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right11(self) -> Self {
         dmap(self, |x| x.rotate_right(11))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right12(self) -> Self {
         dmap(self, |x| x.rotate_right(12))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right16(self) -> Self {
         dmap(self, |x| x.rotate_right(16))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right20(self) -> Self {
         dmap(self, |x| x.rotate_right(20))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right24(self) -> Self {
         dmap(self, |x| x.rotate_right(24))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right25(self) -> Self {
         dmap(self, |x| x.rotate_right(25))
     }
 }
 
 impl RotateEachWord32 for u64x2_generic {
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right7(self) -> Self {
         qmap(self, |x| x.rotate_right(7))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right8(self) -> Self {
         qmap(self, |x| x.rotate_right(8))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right11(self) -> Self {
         qmap(self, |x| x.rotate_right(11))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right12(self) -> Self {
         qmap(self, |x| x.rotate_right(12))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right16(self) -> Self {
         qmap(self, |x| x.rotate_right(16))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right20(self) -> Self {
         qmap(self, |x| x.rotate_right(20))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right24(self) -> Self {
         qmap(self, |x| x.rotate_right(24))
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right25(self) -> Self {
         qmap(self, |x| x.rotate_right(25))
     }
 }
 impl RotateEachWord64 for u64x2_generic {
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right32(self) -> Self {
         qmap(self, |x| x.rotate_right(32))
     }
 }
 
 // workaround for koute/cargo-web#52 (u128::rotate_* broken with cargo web)
+#[inline(always)]
 fn rotate_u128_right(x: u128, i: u32) -> u128 {
     (x >> i) | (x << (128 - i))
 }
@@ -375,41 +393,41 @@ fn test_rotate_u128() {
 }
 
 impl RotateEachWord32 for u128x1_generic {
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right7(self) -> Self {
         Self([rotate_u128_right(self.0[0], 7)])
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right8(self) -> Self {
         Self([rotate_u128_right(self.0[0], 8)])
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right11(self) -> Self {
         Self([rotate_u128_right(self.0[0], 11)])
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right12(self) -> Self {
         Self([rotate_u128_right(self.0[0], 12)])
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right16(self) -> Self {
         Self([rotate_u128_right(self.0[0], 16)])
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right20(self) -> Self {
         Self([rotate_u128_right(self.0[0], 20)])
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right24(self) -> Self {
         Self([rotate_u128_right(self.0[0], 24)])
     }
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right25(self) -> Self {
         Self([rotate_u128_right(self.0[0], 25)])
     }
 }
 impl RotateEachWord64 for u128x1_generic {
-    #[inline]
+    #[inline(always)]
     fn rotate_each_word_right32(self) -> Self {
         Self([rotate_u128_right(self.0[0], 32)])
     }
@@ -428,7 +446,7 @@ impl Machine for GenericMachine {
     type u32x4x4 = u32x4x4_generic;
     type u64x2x4 = u64x2x4_generic;
     type u128x4 = u128x4_generic;
-    #[inline]
+    #[inline(always)]
     unsafe fn instance() -> Self {
         Self
     }
@@ -456,7 +474,7 @@ impl From<u64x2_generic> for vec128_storage {
 impl From<u128x1_generic> for vec128_storage {
     #[inline(always)]
     fn from(o: u128x1_generic) -> Self {
-        Self { o: o.0 }
+        Self { q: q_of_o(o.0[0]) }
     }
 }
 
@@ -475,7 +493,7 @@ impl Store<vec128_storage> for u64x2_generic {
 impl Store<vec128_storage> for u128x1_generic {
     #[inline(always)]
     unsafe fn unpack(s: vec128_storage) -> Self {
-        Self(s.o)
+        Self([o_of_q(s.q); 1])
     }
 }
 
@@ -605,6 +623,22 @@ pub type u32x4x4_generic = x4<u32x4_generic>;
 pub type u64x2x4_generic = x4<u64x2_generic>;
 pub type u128x4_generic = x4<u128x1_generic>;
 
+impl Vector<[u32; 16]> for u32x4x4_generic {
+    fn to_scalars(self) -> [u32; 16] {
+        let [a, b, c, d] = self.0;
+        let a = a.0;
+        let b = b.0;
+        let c = c.0;
+        let d = d.0;
+        [
+            a[0], a[1], a[2], a[3], //
+            b[0], b[1], b[2], b[3], //
+            c[0], c[1], c[2], c[3], //
+            d[0], d[1], d[2], d[3], //
+        ]
+    }
+}
+
 impl MultiLane<[u32; 4]> for u32x4_generic {
     #[inline(always)]
     fn to_lanes(self) -> [u32; 4] {
@@ -745,7 +779,7 @@ impl u128x4<GenericMachine> for u128x4_generic {}
 #[macro_export]
 macro_rules! dispatch {
     ($mach:ident, $MTy:ident, { $([$pub:tt$(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) -> $ret:ty $body:block }) => {
-        #[inline]
+        #[inline(always)]
         $($pub$(($krate))*)* fn $name($($arg: $argty),*) -> $ret {
             let $mach = unsafe { $crate::generic::GenericMachine::instance() };
             #[inline(always)]
@@ -762,7 +796,7 @@ macro_rules! dispatch {
 #[macro_export]
 macro_rules! dispatch_light128 {
     ($mach:ident, $MTy:ident, { $([$pub:tt$(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) -> $ret:ty $body:block }) => {
-        #[inline]
+        #[inline(always)]
         $($pub$(($krate))*)* fn $name($($arg: $argty),*) -> $ret {
             let $mach = unsafe { $crate::generic::GenericMachine::instance() };
             #[inline(always)]
@@ -779,7 +813,7 @@ macro_rules! dispatch_light128 {
 #[macro_export]
 macro_rules! dispatch_light256 {
     ($mach:ident, $MTy:ident, { $([$pub:tt$(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) -> $ret:ty $body:block }) => {
-        #[inline]
+        #[inline(always)]
         $($pub$(($krate))*)* fn $name($($arg: $argty),*) -> $ret {
             let $mach = unsafe { $crate::generic::GenericMachine::instance() };
             #[inline(always)]
@@ -796,7 +830,7 @@ macro_rules! dispatch_light256 {
 #[macro_export]
 macro_rules! dispatch_light512 {
     ($mach:ident, $MTy:ident, { $([$pub:tt$(($krate:tt))*])* fn $name:ident($($arg:ident: $argty:ty),* $(,)*) -> $ret:ty $body:block }) => {
-        #[inline]
+        #[inline(always)]
         $($pub$(($krate))*)* fn $name($($arg: $argty),*) -> $ret {
             let $mach = unsafe { $crate::generic::GenericMachine::instance() };
             #[inline(always)]
diff --git a/vendor/ppv-lite86/src/lib.rs b/vendor/ppv-lite86/src/lib.rs
index 43dc5d869..638552fc2 100644
--- a/vendor/ppv-lite86/src/lib.rs
+++ b/vendor/ppv-lite86/src/lib.rs
@@ -9,14 +9,14 @@ mod soft;
 mod types;
 pub use self::types::*;
 
-#[cfg(all(feature = "simd", target_arch = "x86_64", not(miri)))]
+#[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(feature = "no_simd"), not(miri)))]
 pub mod x86_64;
-#[cfg(all(feature = "simd", target_arch = "x86_64", not(miri)))]
+#[cfg(all(target_arch = "x86_64", target_feature = "sse2", not(feature = "no_simd"), not(miri)))]
 use self::x86_64 as arch;
 
-#[cfg(any(miri, not(all(feature = "simd", any(target_arch = "x86_64")))))]
+#[cfg(any(feature = "no_simd", miri, not(target_arch = "x86_64"), all(target_arch = "x86_64", not(target_feature = "sse2"))))]
 pub mod generic;
-#[cfg(any(miri, not(all(feature = "simd", any(target_arch = "x86_64")))))]
+#[cfg(any(feature = "no_simd", miri, not(target_arch = "x86_64"), all(target_arch = "x86_64", not(target_feature = "sse2"))))]
 use self::generic as arch;
 
 pub use self::arch::{vec128_storage, vec256_storage, vec512_storage};
diff --git a/vendor/ppv-lite86/src/soft.rs b/vendor/ppv-lite86/src/soft.rs
index d12dac528..0ae390c44 100644
--- a/vendor/ppv-lite86/src/soft.rs
+++ b/vendor/ppv-lite86/src/soft.rs
@@ -1,9 +1,9 @@
 //! Implement 256- and 512- bit in terms of 128-bit, for machines without native wide SIMD.
 
-use core::marker::PhantomData;
-use core::ops::*;
 use crate::types::*;
 use crate::{vec128_storage, vec256_storage, vec512_storage};
+use core::marker::PhantomData;
+use core::ops::*;
 
 #[derive(Copy, Clone, Default)]
 #[allow(non_camel_case_types)]
@@ -175,26 +175,50 @@ impl<W: BSwap + Copy, G> BSwap for x2<W, G> {
 impl<W: StoreBytes + BSwap + Copy, G> StoreBytes for x2<W, G> {
     #[inline(always)]
     unsafe fn unsafe_read_le(input: &[u8]) -> Self {
-        let input = input.split_at(16);
+        let input = input.split_at(input.len() / 2);
         x2::new([W::unsafe_read_le(input.0), W::unsafe_read_le(input.1)])
     }
     #[inline(always)]
     unsafe fn unsafe_read_be(input: &[u8]) -> Self {
-        x2::unsafe_read_le(input).bswap()
+        let input = input.split_at(input.len() / 2);
+        x2::new([W::unsafe_read_be(input.0), W::unsafe_read_be(input.1)])
     }
     #[inline(always)]
     fn write_le(self, out: &mut [u8]) {
-        let out = out.split_at_mut(16);
+        let out = out.split_at_mut(out.len() / 2);
         self.0[0].write_le(out.0);
         self.0[1].write_le(out.1);
     }
     #[inline(always)]
     fn write_be(self, out: &mut [u8]) {
-        let out = out.split_at_mut(16);
+        let out = out.split_at_mut(out.len() / 2);
         self.0[0].write_be(out.0);
         self.0[1].write_be(out.1);
     }
 }
+impl<W: Copy + LaneWords4, G: Copy> LaneWords4 for x2<W, G> {
+    #[inline(always)]
+    fn shuffle_lane_words2301(self) -> Self {
+        Self::new([
+            self.0[0].shuffle_lane_words2301(),
+            self.0[1].shuffle_lane_words2301(),
+        ])
+    }
+    #[inline(always)]
+    fn shuffle_lane_words1230(self) -> Self {
+        Self::new([
+            self.0[0].shuffle_lane_words1230(),
+            self.0[1].shuffle_lane_words1230(),
+        ])
+    }
+    #[inline(always)]
+    fn shuffle_lane_words3012(self) -> Self {
+        Self::new([
+            self.0[0].shuffle_lane_words3012(),
+            self.0[1].shuffle_lane_words3012(),
+        ])
+    }
+}
 
 #[derive(Copy, Clone, Default)]
 #[allow(non_camel_case_types)]
@@ -238,7 +262,12 @@ macro_rules! fwd_unop_x4 {
     ($fn:ident) => {
         #[inline(always)]
         fn $fn(self) -> Self {
-            x4([self.0[0].$fn(), self.0[1].$fn(), self.0[2].$fn(), self.0[3].$fn()])
+            x4([
+                self.0[0].$fn(),
+                self.0[1].$fn(),
+                self.0[2].$fn(),
+                self.0[3].$fn(),
+            ])
         }
     };
 }
@@ -305,6 +334,20 @@ impl<W: Copy> Vec4<W> for x4<W> {
         self
     }
 }
+impl<W: Copy> Vec4Ext<W> for x4<W> {
+    #[inline(always)]
+    fn transpose4(a: Self, b: Self, c: Self, d: Self) -> (Self, Self, Self, Self)
+    where
+        Self: Sized,
+    {
+        (
+            x4([a.0[0], b.0[0], c.0[0], d.0[0]]),
+            x4([a.0[1], b.0[1], c.0[1], d.0[1]]),
+            x4([a.0[2], b.0[2], c.0[2], d.0[2]]),
+            x4([a.0[3], b.0[3], c.0[3], d.0[3]]),
+        )
+    }
+}
 impl<W: Copy + Store<vec128_storage>> Store<vec512_storage> for x4<W> {
     #[inline(always)]
     unsafe fn unpack(p: vec512_storage) -> Self {
@@ -363,30 +406,39 @@ impl<W: BSwap + Copy> BSwap for x4<W> {
 impl<W: StoreBytes + BSwap + Copy> StoreBytes for x4<W> {
     #[inline(always)]
     unsafe fn unsafe_read_le(input: &[u8]) -> Self {
+        let n = input.len() / 4;
         x4([
-            W::unsafe_read_le(&input[0..16]),
-            W::unsafe_read_le(&input[16..32]),
-            W::unsafe_read_le(&input[32..48]),
-            W::unsafe_read_le(&input[48..64]),
+            W::unsafe_read_le(&input[..n]),
+            W::unsafe_read_le(&input[n..n * 2]),
+            W::unsafe_read_le(&input[n * 2..n * 3]),
+            W::unsafe_read_le(&input[n * 3..]),
         ])
     }
     #[inline(always)]
     unsafe fn unsafe_read_be(input: &[u8]) -> Self {
-        x4::unsafe_read_le(input).bswap()
+        let n = input.len() / 4;
+        x4([
+            W::unsafe_read_be(&input[..n]),
+            W::unsafe_read_be(&input[n..n * 2]),
+            W::unsafe_read_be(&input[n * 2..n * 3]),
+            W::unsafe_read_be(&input[n * 3..]),
+        ])
     }
     #[inline(always)]
     fn write_le(self, out: &mut [u8]) {
-        self.0[0].write_le(&mut out[0..16]);
-        self.0[1].write_le(&mut out[16..32]);
-        self.0[2].write_le(&mut out[32..48]);
-        self.0[3].write_le(&mut out[48..64]);
+        let n = out.len() / 4;
+        self.0[0].write_le(&mut out[..n]);
+        self.0[1].write_le(&mut out[n..n * 2]);
+        self.0[2].write_le(&mut out[n * 2..n * 3]);
+        self.0[3].write_le(&mut out[n * 3..]);
     }
     #[inline(always)]
     fn write_be(self, out: &mut [u8]) {
-        self.0[0].write_be(&mut out[0..16]);
-        self.0[1].write_be(&mut out[16..32]);
-        self.0[2].write_be(&mut out[32..48]);
-        self.0[3].write_be(&mut out[48..64]);
+        let n = out.len() / 4;
+        self.0[0].write_be(&mut out[..n]);
+        self.0[1].write_be(&mut out[n..n * 2]);
+        self.0[2].write_be(&mut out[n * 2..n * 3]);
+        self.0[3].write_be(&mut out[n * 3..]);
     }
 }
 impl<W: Copy + LaneWords4> LaneWords4 for x4<W> {
diff --git a/vendor/ppv-lite86/src/types.rs b/vendor/ppv-lite86/src/types.rs
index 119b6bb8d..f9f3bf1ce 100644
--- a/vendor/ppv-lite86/src/types.rs
+++ b/vendor/ppv-lite86/src/types.rs
@@ -1,3 +1,4 @@
+#![allow(non_camel_case_types)]
 use core::ops::{Add, AddAssign, BitAnd, BitOr, BitXor, BitXorAssign, Not};
 
 pub trait AndNot {
@@ -44,182 +45,188 @@ pub trait RotateEachWord64 {
 
 pub trait RotateEachWord128 {}
 
-#[allow(non_camel_case_types)]
-mod types {
-    //! Vector type naming scheme:
-    //! uN[xP]xL
-    //! Unsigned; N-bit words * P bits per lane * L lanes
-    //!
-    //! A lane is always 128-bits, chosen because common SIMD architectures treat 128-bit units of
-    //! wide vectors specially (supporting e.g. intra-lane shuffles), and tend to have limited and
-    //! slow inter-lane operations.
+// Vector type naming scheme:
+// uN[xP]xL
+// Unsigned; N-bit words * P bits per lane * L lanes
+//
+// A lane is always 128-bits, chosen because common SIMD architectures treat 128-bit units of
+// wide vectors specially (supporting e.g. intra-lane shuffles), and tend to have limited and
+// slow inter-lane operations.
 
-    use crate::arch::{vec128_storage, vec256_storage, vec512_storage};
-    use crate::{ArithOps, BitOps128, BitOps32, BitOps64, Machine, Store, StoreBytes};
+use crate::arch::{vec128_storage, vec256_storage, vec512_storage};
 
-    pub trait UnsafeFrom<T> {
-        unsafe fn unsafe_from(t: T) -> Self;
-    }
+#[allow(clippy::missing_safety_doc)]
+pub trait UnsafeFrom<T> {
+    unsafe fn unsafe_from(t: T) -> Self;
+}
 
-    /// A vector composed of two elements, which may be words or themselves vectors.
-    pub trait Vec2<W> {
-        fn extract(self, i: u32) -> W;
-        fn insert(self, w: W, i: u32) -> Self;
-    }
+/// A vector composed of two elements, which may be words or themselves vectors.
+pub trait Vec2<W> {
+    fn extract(self, i: u32) -> W;
+    fn insert(self, w: W, i: u32) -> Self;
+}
 
-    /// A vector composed of four elements, which may be words or themselves vectors.
-    pub trait Vec4<W> {
-        fn extract(self, i: u32) -> W;
-        fn insert(self, w: W, i: u32) -> Self;
-    }
+/// A vector composed of four elements, which may be words or themselves vectors.
+pub trait Vec4<W> {
+    fn extract(self, i: u32) -> W;
+    fn insert(self, w: W, i: u32) -> Self;
+}
+/// Vec4 functions which may not be implemented yet for all Vec4 types.
+/// NOTE: functions in this trait may be moved to Vec4 in any patch release. To avoid breakage,
+/// import Vec4Ext only together with Vec4, and don't qualify its methods.
+pub trait Vec4Ext<W> {
+    fn transpose4(a: Self, b: Self, c: Self, d: Self) -> (Self, Self, Self, Self)
+    where
+        Self: Sized;
+}
+pub trait Vector<T> {
+    fn to_scalars(self) -> T;
+}
 
-    // TODO: multiples of 4 should inherit this
-    /// A vector composed of four words; depending on their size, operations may cross lanes.
-    pub trait Words4 {
-        fn shuffle1230(self) -> Self;
-        fn shuffle2301(self) -> Self;
-        fn shuffle3012(self) -> Self;
-    }
+// TODO: multiples of 4 should inherit this
+/// A vector composed of four words; depending on their size, operations may cross lanes.
+pub trait Words4 {
+    fn shuffle1230(self) -> Self;
+    fn shuffle2301(self) -> Self;
+    fn shuffle3012(self) -> Self;
+}
 
-    /// A vector composed one or more lanes each composed of four words.
-    pub trait LaneWords4 {
-        fn shuffle_lane_words1230(self) -> Self;
-        fn shuffle_lane_words2301(self) -> Self;
-        fn shuffle_lane_words3012(self) -> Self;
-    }
+/// A vector composed one or more lanes each composed of four words.
+pub trait LaneWords4 {
+    fn shuffle_lane_words1230(self) -> Self;
+    fn shuffle_lane_words2301(self) -> Self;
+    fn shuffle_lane_words3012(self) -> Self;
+}
 
-    // TODO: make this a part of BitOps
-    /// Exchange neigboring ranges of bits of the specified size
-    pub trait Swap64 {
-        fn swap1(self) -> Self;
-        fn swap2(self) -> Self;
-        fn swap4(self) -> Self;
-        fn swap8(self) -> Self;
-        fn swap16(self) -> Self;
-        fn swap32(self) -> Self;
-        fn swap64(self) -> Self;
-    }
+// TODO: make this a part of BitOps
+/// Exchange neigboring ranges of bits of the specified size
+pub trait Swap64 {
+    fn swap1(self) -> Self;
+    fn swap2(self) -> Self;
+    fn swap4(self) -> Self;
+    fn swap8(self) -> Self;
+    fn swap16(self) -> Self;
+    fn swap32(self) -> Self;
+    fn swap64(self) -> Self;
+}
 
-    pub trait u32x4<M: Machine>:
-        BitOps32
-        + Store<vec128_storage>
-        + ArithOps
-        + Vec4<u32>
-        + Words4
-        + LaneWords4
-        + StoreBytes
-        + MultiLane<[u32; 4]>
-        + Into<vec128_storage>
-    {
+pub trait u32x4<M: Machine>:
+    BitOps32
+    + Store<vec128_storage>
+    + ArithOps
+    + Vec4<u32>
+    + Words4
+    + LaneWords4
+    + StoreBytes
+    + MultiLane<[u32; 4]>
+    + Into<vec128_storage>
+{
 }
-    pub trait u64x2<M: Machine>:
-        BitOps64
-        + Store<vec128_storage>
-        + ArithOps
-        + Vec2<u64>
-        + MultiLane<[u64; 2]>
-        + Into<vec128_storage>
-    {
+pub trait u64x2<M: Machine>:
+    BitOps64 + Store<vec128_storage> + ArithOps + Vec2<u64> + MultiLane<[u64; 2]> + Into<vec128_storage>
+{
 }
-    pub trait u128x1<M: Machine>:
-        BitOps128 + Store<vec128_storage> + Swap64 + MultiLane<[u128; 1]> + Into<vec128_storage>
-    {
+pub trait u128x1<M: Machine>:
+    BitOps128 + Store<vec128_storage> + Swap64 + MultiLane<[u128; 1]> + Into<vec128_storage>
+{
 }
 
-    pub trait u32x4x2<M: Machine>:
-        BitOps32
-        + Store<vec256_storage>
-        + Vec2<M::u32x4>
-        + MultiLane<[M::u32x4; 2]>
-        + ArithOps
-        + Into<vec256_storage>
-    {
+pub trait u32x4x2<M: Machine>:
+    BitOps32
+    + Store<vec256_storage>
+    + Vec2<M::u32x4>
+    + MultiLane<[M::u32x4; 2]>
+    + ArithOps
+    + Into<vec256_storage>
+    + StoreBytes
+{
 }
-    pub trait u64x2x2<M: Machine>:
-        BitOps64
-        + Store<vec256_storage>
-        + Vec2<M::u64x2>
-        + MultiLane<[M::u64x2; 2]>
-        + ArithOps
-        + StoreBytes
-        + Into<vec256_storage>
-    {
+pub trait u64x2x2<M: Machine>:
+    BitOps64
+    + Store<vec256_storage>
+    + Vec2<M::u64x2>
+    + MultiLane<[M::u64x2; 2]>
+    + ArithOps
+    + StoreBytes
+    + Into<vec256_storage>
+{
 }
-    pub trait u64x4<M: Machine>:
-        BitOps64
-        + Store<vec256_storage>
-        + Vec4<u64>
-        + MultiLane<[u64; 4]>
-        + ArithOps
-        + Words4
-        + StoreBytes
-        + Into<vec256_storage>
-    {
+pub trait u64x4<M: Machine>:
+    BitOps64
+    + Store<vec256_storage>
+    + Vec4<u64>
+    + MultiLane<[u64; 4]>
+    + ArithOps
+    + Words4
+    + StoreBytes
+    + Into<vec256_storage>
+{
 }
-    pub trait u128x2<M: Machine>:
-        BitOps128
-        + Store<vec256_storage>
-        + Vec2<M::u128x1>
-        + MultiLane<[M::u128x1; 2]>
-        + Swap64
-        + Into<vec256_storage>
-    {
+pub trait u128x2<M: Machine>:
+    BitOps128
+    + Store<vec256_storage>
+    + Vec2<M::u128x1>
+    + MultiLane<[M::u128x1; 2]>
+    + Swap64
+    + Into<vec256_storage>
+{
 }
 
-    pub trait u32x4x4<M: Machine>:
-        BitOps32
-        + Store<vec512_storage>
-        + Vec4<M::u32x4>
-        + MultiLane<[M::u32x4; 4]>
-        + ArithOps
-        + LaneWords4
-        + Into<vec512_storage>
-    {
+pub trait u32x4x4<M: Machine>:
+    BitOps32
+    + Store<vec512_storage>
+    + Vec4<M::u32x4>
+    + Vec4Ext<M::u32x4>
+    + Vector<[u32; 16]>
+    + MultiLane<[M::u32x4; 4]>
+    + ArithOps
+    + LaneWords4
+    + Into<vec512_storage>
+    + StoreBytes
+{
 }
-    pub trait u64x2x4<M: Machine>:
-        BitOps64
-        + Store<vec512_storage>
-        + Vec4<M::u64x2>
-        + MultiLane<[M::u64x2; 4]>
-        + ArithOps
-        + Into<vec512_storage>
-    {
+pub trait u64x2x4<M: Machine>:
+    BitOps64
+    + Store<vec512_storage>
+    + Vec4<M::u64x2>
+    + MultiLane<[M::u64x2; 4]>
+    + ArithOps
+    + Into<vec512_storage>
+{
 }
-    // TODO: Words4
-    pub trait u128x4<M: Machine>:
-        BitOps128
-        + Store<vec512_storage>
-        + Vec4<M::u128x1>
-        + MultiLane<[M::u128x1; 4]>
-        + Swap64
-        + Into<vec512_storage>
-    {
+// TODO: Words4
+pub trait u128x4<M: Machine>:
+    BitOps128
+    + Store<vec512_storage>
+    + Vec4<M::u128x1>
+    + MultiLane<[M::u128x1; 4]>
+    + Swap64
+    + Into<vec512_storage>
+{
 }
 
-    /// A vector composed of multiple 128-bit lanes.
-    pub trait MultiLane<Lanes> {
-        /// Split a multi-lane vector into single-lane vectors.
-        fn to_lanes(self) -> Lanes;
-        /// Build a multi-lane vector from individual lanes.
-        fn from_lanes(lanes: Lanes) -> Self;
-    }
+/// A vector composed of multiple 128-bit lanes.
+pub trait MultiLane<Lanes> {
+    /// Split a multi-lane vector into single-lane vectors.
+    fn to_lanes(self) -> Lanes;
+    /// Build a multi-lane vector from individual lanes.
+    fn from_lanes(lanes: Lanes) -> Self;
+}
 
-    /// Combine single vectors into a multi-lane vector.
-    pub trait VZip<V> {
-        fn vzip(self) -> V;
-    }
+/// Combine single vectors into a multi-lane vector.
+pub trait VZip<V> {
+    fn vzip(self) -> V;
+}
 
-    impl<V, T> VZip<V> for T
-    where
-        V: MultiLane<T>,
-    {
-        #[inline(always)]
-        fn vzip(self) -> V {
-            V::from_lanes(self)
-        }
+impl<V, T> VZip<V> for T
+where
+    V: MultiLane<T>,
+{
+    #[inline(always)]
+    fn vzip(self) -> V {
+        V::from_lanes(self)
     }
 }
-pub use self::types::*;
 
 pub trait Machine: Sized + Copy {
     type u32x4: u32x4<Self>;
@@ -264,15 +271,27 @@ pub trait Machine: Sized + Copy {
         unsafe { V::unsafe_read_be(input) }
     }
 
+    /// # Safety
+    /// Caller must ensure the type of Self is appropriate for the hardware of the execution
+    /// environment.
     unsafe fn instance() -> Self;
 }
 
 pub trait Store<S> {
+    /// # Safety
+    /// Caller must ensure the type of Self is appropriate for the hardware of the execution
+    /// environment.
     unsafe fn unpack(p: S) -> Self;
 }
 
 pub trait StoreBytes {
+    /// # Safety
+    /// Caller must ensure the type of Self is appropriate for the hardware of the execution
+    /// environment.
     unsafe fn unsafe_read_le(input: &[u8]) -> Self;
+    /// # Safety
+    /// Caller must ensure the type of Self is appropriate for the hardware of the execution
+    /// environment.
     unsafe fn unsafe_read_be(input: &[u8]) -> Self;
     fn write_le(self, out: &mut [u8]);
     fn write_be(self, out: &mut [u8]);
diff --git a/vendor/ppv-lite86/src/x86_64/mod.rs b/vendor/ppv-lite86/src/x86_64/mod.rs
index ecf184f36..937732da3 100644
--- a/vendor/ppv-lite86/src/x86_64/mod.rs
+++ b/vendor/ppv-lite86/src/x86_64/mod.rs
@@ -1,7 +1,7 @@
 // crate minimums: sse2, x86_64
 
-use core::arch::x86_64::{__m128i, __m256i};
 use crate::types::*;
+use core::arch::x86_64::{__m128i, __m256i};
 
 mod sse2;
 
@@ -79,7 +79,7 @@ where
     type u64x2 = sse2::u64x2_sse2<YesS3, YesS4, NI>;
     type u128x1 = sse2::u128x1_sse2<YesS3, YesS4, NI>;
 
-    type u32x4x2 = sse2::u32x4x2_sse2<YesS3, YesS4, NI>;
+    type u32x4x2 = sse2::avx2::u32x4x2_avx2<NI>;
     type u64x2x2 = sse2::u64x2x2_sse2<YesS3, YesS4, NI>;
     type u64x4 = sse2::u64x4_sse2<YesS3, YesS4, NI>;
     type u128x2 = sse2::u128x2_sse2<YesS3, YesS4, NI>;
@@ -119,16 +119,16 @@ impl Store<vec128_storage> for vec128_storage {
         p
     }
 }
-impl<'a> Into<&'a [u32; 4]> for &'a vec128_storage {
+impl<'a> From<&'a vec128_storage> for &'a [u32; 4] {
     #[inline(always)]
-    fn into(self) -> &'a [u32; 4] {
-        unsafe { &self.u32x4 }
+    fn from(x: &'a vec128_storage) -> Self {
+        unsafe { &x.u32x4 }
     }
 }
-impl Into<vec128_storage> for [u32; 4] {
+impl From<[u32; 4]> for vec128_storage {
     #[inline(always)]
-    fn into(self) -> vec128_storage {
-        vec128_storage { u32x4: self }
+    fn from(u32x4: [u32; 4]) -> Self {
+        vec128_storage { u32x4 }
     }
 }
 impl Default for vec128_storage {
@@ -154,10 +154,10 @@ pub union vec256_storage {
     sse2: [vec128_storage; 2],
     avx: __m256i,
 }
-impl Into<vec256_storage> for [u64; 4] {
+impl From<[u64; 4]> for vec256_storage {
     #[inline(always)]
-    fn into(self) -> vec256_storage {
-        vec256_storage { u64x4: self }
+    fn from(u64x4: [u64; 4]) -> Self {
+        vec256_storage { u64x4 }
     }
 }
 impl Default for vec256_storage {
@@ -167,9 +167,11 @@ impl Default for vec256_storage {
     }
 }
 impl vec256_storage {
+    #[inline(always)]
     pub fn new128(xs: [vec128_storage; 2]) -> Self {
         Self { sse2: xs }
     }
+    #[inline(always)]
     pub fn split128(self) -> [vec128_storage; 2] {
         unsafe { self.sse2 }
     }
@@ -200,9 +202,11 @@ impl Default for vec512_storage {
     }
 }
 impl vec512_storage {
+    #[inline(always)]
     pub fn new128(xs: [vec128_storage; 4]) -> Self {
         Self { sse2: xs }
     }
+    #[inline(always)]
     pub fn split128(self) -> [vec128_storage; 4] {
         unsafe { self.sse2 }
     }
@@ -217,10 +221,10 @@ impl PartialEq for vec512_storage {
 
 macro_rules! impl_into {
     ($storage:ident, $array:ty, $name:ident) => {
-        impl Into<$array> for $storage {
+        impl From<$storage> for $array {
             #[inline(always)]
-            fn into(self) -> $array {
-                unsafe { self.$name }
+            fn from(vec: $storage) -> Self {
+                unsafe { vec.$name }
             }
         }
     };
diff --git a/vendor/ppv-lite86/src/x86_64/sse2.rs b/vendor/ppv-lite86/src/x86_64/sse2.rs
index 60e7681c3..97197a436 100644
--- a/vendor/ppv-lite86/src/x86_64/sse2.rs
+++ b/vendor/ppv-lite86/src/x86_64/sse2.rs
@@ -166,49 +166,44 @@ macro_rules! impl_bitops128 {
 
 macro_rules! rotr_32_s3 {
     ($name:ident, $k0:expr, $k1:expr) => {
-    #[inline(always)]
-    fn $name(self) -> Self {
-        Self::new(unsafe {
-                _mm_shuffle_epi8(
-                    self.x,
-                    _mm_set_epi64x($k0, $k1),
-                )
-            })
+        #[inline(always)]
+        fn $name(self) -> Self {
+            Self::new(unsafe { _mm_shuffle_epi8(self.x, _mm_set_epi64x($k0, $k1)) })
         }
     };
 }
 macro_rules! rotr_32 {
     ($name:ident, $i:expr) => {
-    #[inline(always)]
-    fn $name(self) -> Self {
-        Self::new(unsafe {
-            _mm_or_si128(
-                _mm_srli_epi32(self.x, $i as i32),
-                _mm_slli_epi32(self.x, 32 - $i as i32),
-            )
-        })
-    }
+        #[inline(always)]
+        fn $name(self) -> Self {
+            Self::new(unsafe {
+                _mm_or_si128(
+                    _mm_srli_epi32(self.x, $i as i32),
+                    _mm_slli_epi32(self.x, 32 - $i as i32),
+                )
+            })
+        }
     };
 }
 impl<S4: Copy, NI: Copy> RotateEachWord32 for u32x4_sse2<YesS3, S4, NI> {
     rotr_32!(rotate_each_word_right7, 7);
     rotr_32_s3!(
         rotate_each_word_right8,
-        0x0c0f0e0d_080b0a09,
-        0x04070605_00030201
+        0x0c0f_0e0d_080b_0a09,
+        0x0407_0605_0003_0201
     );
     rotr_32!(rotate_each_word_right11, 11);
     rotr_32!(rotate_each_word_right12, 12);
     rotr_32_s3!(
         rotate_each_word_right16,
-        0x0d0c0f0e_09080b0a,
-        0x05040706_01000302
+        0x0d0c_0f0e_0908_0b0a,
+        0x0504_0706_0100_0302
     );
     rotr_32!(rotate_each_word_right20, 20);
     rotr_32_s3!(
         rotate_each_word_right24,
-        0x0e0d0c0f_0a09080b,
-        0x06050407_02010003
+        0x0e0d_0c0f_0a09_080b,
+        0x0605_0407_0201_0003
     );
     rotr_32!(rotate_each_word_right25, 25);
 }
@@ -228,28 +223,23 @@ impl<S4: Copy, NI: Copy> RotateEachWord32 for u32x4_sse2<NoS3, S4, NI> {
 
 macro_rules! rotr_64_s3 {
     ($name:ident, $k0:expr, $k1:expr) => {
-    #[inline(always)]
-    fn $name(self) -> Self {
-        Self::new(unsafe {
-                _mm_shuffle_epi8(
-                    self.x,
-                    _mm_set_epi64x($k0, $k1),
-                )
-            })
+        #[inline(always)]
+        fn $name(self) -> Self {
+            Self::new(unsafe { _mm_shuffle_epi8(self.x, _mm_set_epi64x($k0, $k1)) })
         }
     };
 }
 macro_rules! rotr_64 {
     ($name:ident, $i:expr) => {
-    #[inline(always)]
-    fn $name(self) -> Self {
-        Self::new(unsafe {
-            _mm_or_si128(
-                _mm_srli_epi64(self.x, $i as i32),
-                _mm_slli_epi64(self.x, 64 - $i as i32),
-            )
-        })
-    }
+        #[inline(always)]
+        fn $name(self) -> Self {
+            Self::new(unsafe {
+                _mm_or_si128(
+                    _mm_srli_epi64(self.x, $i as i32),
+                    _mm_slli_epi64(self.x, 64 - $i as i32),
+                )
+            })
+        }
     };
 }
 impl<S4: Copy, NI: Copy> RotateEachWord32 for u64x2_sse2<YesS3, S4, NI> {
@@ -296,15 +286,15 @@ impl<S3: Copy, S4: Copy, NI: Copy> RotateEachWord64 for u64x2_sse2<S3, S4, NI> {
 
 macro_rules! rotr_128 {
     ($name:ident, $i:expr) => {
-    #[inline(always)]
-    fn $name(self) -> Self {
-        Self::new(unsafe {
-            _mm_or_si128(
-                _mm_srli_si128(self.x, $i as i32),
-                _mm_slli_si128(self.x, 128 - $i as i32),
-            )
-        })
-    }
+        #[inline(always)]
+        fn $name(self) -> Self {
+            Self::new(unsafe {
+                _mm_or_si128(
+                    _mm_srli_si128(self.x, $i as i32),
+                    _mm_slli_si128(self.x, 128 - $i as i32),
+                )
+            })
+        }
     };
 }
 // TODO: completely unoptimized
@@ -411,7 +401,7 @@ impl<S3, S4, NI> MultiLane<[u128; 1]> for u128x1_sse2<S3, S4, NI> {
     }
     #[inline(always)]
     fn from_lanes(xs: [u128; 1]) -> Self {
-        unimplemented!()
+        unimplemented!("{:?}", xs)
     }
 }
 
@@ -780,7 +770,7 @@ impl<S4, NI> BSwap for u128x1_sse2<YesS3, S4, NI> {
 impl<S4, NI> BSwap for u128x1_sse2<NoS3, S4, NI> {
     #[inline(always)]
     fn bswap(self) -> Self {
-        Self::new(unsafe { unimplemented!() })
+        unimplemented!()
     }
 }
 
@@ -890,6 +880,13 @@ pub type u64x2x4_sse2<S3, S4, NI> = x4<u64x2_sse2<S3, S4, NI>>;
 #[allow(non_camel_case_types)]
 pub type u128x4_sse2<S3, S4, NI> = x4<u128x1_sse2<S3, S4, NI>>;
 
+impl<S3, S4, NI> Vector<[u32; 16]> for u32x4x4_sse2<S3, S4, NI> {
+    #[inline(always)]
+    fn to_scalars(self) -> [u32; 16] {
+        unsafe { core::mem::transmute(self) }
+    }
+}
+
 impl<S3: Copy, S4: Copy, NI: Copy> u32x4x2<Machine86<S3, S4, NI>> for u32x4x2_sse2<S3, S4, NI>
 where
     u32x4_sse2<S3, S4, NI>: RotateEachWord32 + BSwap,
@@ -993,6 +990,8 @@ where
     Machine86<S3, S4, NI>: Machine,
     u32x4x4_sse2<S3, S4, NI>: MultiLane<[<Machine86<S3, S4, NI> as Machine>::u32x4; 4]>,
     u32x4x4_sse2<S3, S4, NI>: Vec4<<Machine86<S3, S4, NI> as Machine>::u32x4>,
+    u32x4x4_sse2<S3, S4, NI>: Vec4Ext<<Machine86<S3, S4, NI> as Machine>::u32x4>,
+    u32x4x4_sse2<S3, S4, NI>: Vector<[u32; 16]>,
 {
 }
 impl<S3: Copy, S4: Copy, NI: Copy> u64x2x4<Machine86<S3, S4, NI>> for u64x2x4_sse2<S3, S4, NI>
@@ -1014,14 +1013,6 @@ where
 {
 }
 
-impl<NI: Copy> u32x4x4<Avx2Machine<NI>> for u32x4x4_sse2<YesS3, YesS4, NI>
-where
-    u32x4_sse2<YesS3, YesS4, NI>: RotateEachWord32 + BSwap,
-    Avx2Machine<NI>: Machine,
-    u32x4x4_sse2<YesS3, YesS4, NI>: MultiLane<[<Avx2Machine<NI> as Machine>::u32x4; 4]>,
-    u32x4x4_sse2<YesS3, YesS4, NI>: Vec4<<Avx2Machine<NI> as Machine>::u32x4>,
-{
-}
 impl<NI: Copy> u64x2x4<Avx2Machine<NI>> for u64x2x4_sse2<YesS3, YesS4, NI>
 where
     u64x2_sse2<YesS3, YesS4, NI>: RotateEachWord64 + RotateEachWord32 + BSwap,
@@ -1078,6 +1069,7 @@ impl<W: PartialEq, G> PartialEq for x2<W, G> {
     }
 }
 
+#[allow(unused)]
 #[inline(always)]
 unsafe fn eq128_s4(x: __m128i, y: __m128i) -> bool {
     let q = _mm_shuffle_epi32(_mm_cmpeq_epi64(x, y), 0b1100_0110);
@@ -1383,65 +1375,78 @@ mod test {
 
 pub mod avx2 {
     #![allow(non_camel_case_types)]
-    use crate::soft::x4;
+    use crate::soft::{x2, x4};
     use crate::types::*;
-    use crate::x86_64::sse2::{u128x1_sse2, u32x4_sse2};
+    use crate::x86_64::sse2::{u128x1_sse2, u32x4_sse2, G0};
     use crate::x86_64::{vec256_storage, vec512_storage, Avx2Machine, YesS3, YesS4};
     use core::arch::x86_64::*;
     use core::marker::PhantomData;
     use core::ops::*;
 
     #[derive(Copy, Clone)]
-    pub struct u32x4x4_avx2<NI> {
-        x: [__m256i; 2],
+    pub struct u32x4x2_avx2<NI> {
+        x: __m256i,
         ni: PhantomData<NI>,
     }
 
-    impl<NI> u32x4x4_avx2<NI> {
+    impl<NI> u32x4x2_avx2<NI> {
         #[inline(always)]
-        fn new(x: [__m256i; 2]) -> Self {
+        fn new(x: __m256i) -> Self {
             Self { x, ni: PhantomData }
         }
     }
 
-    impl<NI> u32x4x4<Avx2Machine<NI>> for u32x4x4_avx2<NI> where NI: Copy {}
-    impl<NI> Store<vec512_storage> for u32x4x4_avx2<NI> {
+    impl<NI> u32x4x2<Avx2Machine<NI>> for u32x4x2_avx2<NI> where NI: Copy {}
+    impl<NI> Store<vec256_storage> for u32x4x2_avx2<NI> {
         #[inline(always)]
-        unsafe fn unpack(p: vec512_storage) -> Self {
-            Self::new([p.avx[0].avx, p.avx[1].avx])
+        unsafe fn unpack(p: vec256_storage) -> Self {
+            Self::new(p.avx)
         }
     }
-    impl<NI> MultiLane<[u32x4_sse2<YesS3, YesS4, NI>; 4]> for u32x4x4_avx2<NI> {
+    impl<NI> StoreBytes for u32x4x2_avx2<NI> {
         #[inline(always)]
-        fn to_lanes(self) -> [u32x4_sse2<YesS3, YesS4, NI>; 4] {
+        unsafe fn unsafe_read_le(input: &[u8]) -> Self {
+            assert_eq!(input.len(), 32);
+            Self::new(_mm256_loadu_si256(input.as_ptr() as *const _))
+        }
+        #[inline(always)]
+        unsafe fn unsafe_read_be(input: &[u8]) -> Self {
+            Self::unsafe_read_le(input).bswap()
+        }
+        #[inline(always)]
+        fn write_le(self, out: &mut [u8]) {
             unsafe {
-                [
-                    u32x4_sse2::new(_mm256_extracti128_si256(self.x[0], 0)),
-                    u32x4_sse2::new(_mm256_extracti128_si256(self.x[0], 1)),
-                    u32x4_sse2::new(_mm256_extracti128_si256(self.x[1], 0)),
-                    u32x4_sse2::new(_mm256_extracti128_si256(self.x[1], 1)),
-                ]
+                assert_eq!(out.len(), 32);
+                _mm256_storeu_si256(out.as_mut_ptr() as *mut _, self.x)
             }
         }
         #[inline(always)]
-        fn from_lanes(x: [u32x4_sse2<YesS3, YesS4, NI>; 4]) -> Self {
-            Self::new(unsafe {
+        fn write_be(self, out: &mut [u8]) {
+            self.bswap().write_le(out)
+        }
+    }
+    impl<NI> MultiLane<[u32x4_sse2<YesS3, YesS4, NI>; 2]> for u32x4x2_avx2<NI> {
+        #[inline(always)]
+        fn to_lanes(self) -> [u32x4_sse2<YesS3, YesS4, NI>; 2] {
+            unsafe {
                 [
-                    _mm256_setr_m128i(x[0].x, x[1].x),
-                    _mm256_setr_m128i(x[2].x, x[3].x),
+                    u32x4_sse2::new(_mm256_extracti128_si256(self.x, 0)),
+                    u32x4_sse2::new(_mm256_extracti128_si256(self.x, 1)),
                 ]
-            })
+            }
+        }
+        #[inline(always)]
+        fn from_lanes(x: [u32x4_sse2<YesS3, YesS4, NI>; 2]) -> Self {
+            Self::new(unsafe { _mm256_setr_m128i(x[0].x, x[1].x) })
         }
     }
-    impl<NI> Vec4<u32x4_sse2<YesS3, YesS4, NI>> for u32x4x4_avx2<NI> {
+    impl<NI> Vec2<u32x4_sse2<YesS3, YesS4, NI>> for u32x4x2_avx2<NI> {
         #[inline(always)]
         fn extract(self, i: u32) -> u32x4_sse2<YesS3, YesS4, NI> {
             unsafe {
                 match i {
-                    0 => u32x4_sse2::new(_mm256_extracti128_si256(self.x[0], 0)),
-                    1 => u32x4_sse2::new(_mm256_extracti128_si256(self.x[0], 1)),
-                    2 => u32x4_sse2::new(_mm256_extracti128_si256(self.x[1], 0)),
-                    3 => u32x4_sse2::new(_mm256_extracti128_si256(self.x[1], 1)),
+                    0 => u32x4_sse2::new(_mm256_extracti128_si256(self.x, 0)),
+                    1 => u32x4_sse2::new(_mm256_extracti128_si256(self.x, 1)),
                     _ => panic!(),
                 }
             }
@@ -1450,61 +1455,21 @@ pub mod avx2 {
         fn insert(self, w: u32x4_sse2<YesS3, YesS4, NI>, i: u32) -> Self {
             Self::new(unsafe {
                 match i {
-                    0 => [_mm256_inserti128_si256(self.x[0], w.x, 0), self.x[1]],
-                    1 => [_mm256_inserti128_si256(self.x[0], w.x, 1), self.x[1]],
-                    2 => [self.x[0], _mm256_inserti128_si256(self.x[1], w.x, 0)],
-                    3 => [self.x[0], _mm256_inserti128_si256(self.x[1], w.x, 1)],
+                    0 => _mm256_inserti128_si256(self.x, w.x, 0),
+                    1 => _mm256_inserti128_si256(self.x, w.x, 1),
                     _ => panic!(),
                 }
             })
         }
     }
-    impl<NI> LaneWords4 for u32x4x4_avx2<NI> {
-        #[inline(always)]
-        fn shuffle_lane_words1230(self) -> Self {
-            Self::new(unsafe {
-                [
-                    _mm256_shuffle_epi32(self.x[0], 0b1001_0011),
-                    _mm256_shuffle_epi32(self.x[1], 0b1001_0011),
-                ]
-            })
-        }
-        #[inline(always)]
-        fn shuffle_lane_words2301(self) -> Self {
-            Self::new(unsafe {
-                [
-                    _mm256_shuffle_epi32(self.x[0], 0b0100_1110),
-                    _mm256_shuffle_epi32(self.x[1], 0b0100_1110),
-                ]
-            })
-        }
-        #[inline(always)]
-        fn shuffle_lane_words3012(self) -> Self {
-            Self::new(unsafe {
-                [
-                    _mm256_shuffle_epi32(self.x[0], 0b0011_1001),
-                    _mm256_shuffle_epi32(self.x[1], 0b0011_1001),
-                ]
-            })
-        }
-    }
-    impl<NI> BitOps32 for u32x4x4_avx2<NI> where NI: Copy {}
-    impl<NI> ArithOps for u32x4x4_avx2<NI> where NI: Copy {}
+    impl<NI> BitOps32 for u32x4x2_avx2<NI> where NI: Copy {}
+    impl<NI> ArithOps for u32x4x2_avx2<NI> where NI: Copy {}
     macro_rules! shuf_lane_bytes {
         ($name:ident, $k0:expr, $k1:expr) => {
-        #[inline(always)]
-        fn $name(self) -> Self {
-            Self::new(unsafe {
-                [
-                    _mm256_shuffle_epi8(
-                        self.x[0],
-                        _mm256_set_epi64x($k0, $k1, $k0, $k1),
-                    ),
-                    _mm256_shuffle_epi8(
-                        self.x[1],
-                        _mm256_set_epi64x($k0, $k1, $k0, $k1),
-                    )
-                ]
+            #[inline(always)]
+            fn $name(self) -> Self {
+                Self::new(unsafe {
+                    _mm256_shuffle_epi8(self.x, _mm256_set_epi64x($k0, $k1, $k0, $k1))
                 })
             }
         };
@@ -1514,52 +1479,41 @@ pub mod avx2 {
             #[inline(always)]
             fn $name(self) -> Self {
                 Self::new(unsafe {
-                    [
-                        _mm256_or_si256(
-                            _mm256_srli_epi32(self.x[0], $i as i32),
-                            _mm256_slli_epi32(self.x[0], 32 - $i as i32),
-                        ),
-                        _mm256_or_si256(
-                            _mm256_srli_epi32(self.x[1], $i as i32),
-                            _mm256_slli_epi32(self.x[1], 32 - $i as i32),
-                        )
-                    ]
+                    _mm256_or_si256(
+                        _mm256_srli_epi32(self.x, $i as i32),
+                        _mm256_slli_epi32(self.x, 32 - $i as i32),
+                    )
                 })
             }
         };
     }
-    impl<NI: Copy> RotateEachWord32 for u32x4x4_avx2<NI> {
+    impl<NI: Copy> RotateEachWord32 for u32x4x2_avx2<NI> {
         rotr_32!(rotate_each_word_right7, 7);
         shuf_lane_bytes!(
             rotate_each_word_right8,
-            0x0c0f0e0d_080b0a09,
-            0x04070605_00030201
+            0x0c0f_0e0d_080b_0a09,
+            0x0407_0605_0003_0201
         );
         rotr_32!(rotate_each_word_right11, 11);
         rotr_32!(rotate_each_word_right12, 12);
         shuf_lane_bytes!(
             rotate_each_word_right16,
-            0x0d0c0f0e_09080b0a,
-            0x05040706_01000302
+            0x0d0c_0f0e_0908_0b0a,
+            0x0504_0706_0100_0302
         );
         rotr_32!(rotate_each_word_right20, 20);
         shuf_lane_bytes!(
             rotate_each_word_right24,
-            0x0e0d0c0f_0a09080b,
-            0x06050407_02010003
+            0x0e0d_0c0f_0a09_080b,
+            0x0605_0407_0201_0003
         );
         rotr_32!(rotate_each_word_right25, 25);
     }
-    impl<NI> BitOps0 for u32x4x4_avx2<NI> where NI: Copy {}
-    impl<NI> From<u32x4x4_avx2<NI>> for vec512_storage {
+    impl<NI> BitOps0 for u32x4x2_avx2<NI> where NI: Copy {}
+    impl<NI> From<u32x4x2_avx2<NI>> for vec256_storage {
         #[inline(always)]
-        fn from(x: u32x4x4_avx2<NI>) -> Self {
-            Self {
-                avx: [
-                    vec256_storage { avx: x.x[0] },
-                    vec256_storage { avx: x.x[1] },
-                ],
-            }
+        fn from(x: u32x4x2_avx2<NI>) -> Self {
+            Self { avx: x.x }
         }
     }
 
@@ -1576,55 +1530,172 @@ pub mod avx2 {
             }
         };
     }
-    impl_assign!(u32x4x4_avx2, BitXorAssign, bitxor_assign, bitxor);
-    impl_assign!(u32x4x4_avx2, BitOrAssign, bitor_assign, bitor);
-    impl_assign!(u32x4x4_avx2, BitAndAssign, bitand_assign, bitand);
-    impl_assign!(u32x4x4_avx2, AddAssign, add_assign, add);
+    impl_assign!(u32x4x2_avx2, BitXorAssign, bitxor_assign, bitxor);
+    impl_assign!(u32x4x2_avx2, BitOrAssign, bitor_assign, bitor);
+    impl_assign!(u32x4x2_avx2, BitAndAssign, bitand_assign, bitand);
+    impl_assign!(u32x4x2_avx2, AddAssign, add_assign, add);
 
-    macro_rules! impl_bitop_x2 {
+    macro_rules! impl_bitop {
         ($vec:ident, $Op:ident, $op_fn:ident, $impl_fn:ident) => {
             impl<NI> $Op for $vec<NI> {
                 type Output = Self;
                 #[inline(always)]
                 fn $op_fn(self, rhs: Self) -> Self::Output {
-                    Self::new(unsafe {
-                        [$impl_fn(self.x[0], rhs.x[0]), $impl_fn(self.x[1], rhs.x[1])]
-                    })
+                    Self::new(unsafe { $impl_fn(self.x, rhs.x) })
                 }
             }
         };
     }
-    impl_bitop_x2!(u32x4x4_avx2, BitXor, bitxor, _mm256_xor_si256);
-    impl_bitop_x2!(u32x4x4_avx2, BitOr, bitor, _mm256_or_si256);
-    impl_bitop_x2!(u32x4x4_avx2, BitAnd, bitand, _mm256_and_si256);
-    impl_bitop_x2!(u32x4x4_avx2, AndNot, andnot, _mm256_andnot_si256);
-    impl_bitop_x2!(u32x4x4_avx2, Add, add, _mm256_add_epi32);
+    impl_bitop!(u32x4x2_avx2, BitXor, bitxor, _mm256_xor_si256);
+    impl_bitop!(u32x4x2_avx2, BitOr, bitor, _mm256_or_si256);
+    impl_bitop!(u32x4x2_avx2, BitAnd, bitand, _mm256_and_si256);
+    impl_bitop!(u32x4x2_avx2, AndNot, andnot, _mm256_andnot_si256);
+    impl_bitop!(u32x4x2_avx2, Add, add, _mm256_add_epi32);
 
-    impl<NI> Not for u32x4x4_avx2<NI> {
+    impl<NI> Not for u32x4x2_avx2<NI> {
         type Output = Self;
         #[inline(always)]
         fn not(self) -> Self::Output {
             unsafe {
                 let f = _mm256_set1_epi8(-0x7f);
-                Self::new([f, f]) ^ self
+                Self::new(f) ^ self
             }
         }
     }
 
-    impl<NI> BSwap for u32x4x4_avx2<NI> {
+    impl<NI> BSwap for u32x4x2_avx2<NI> {
         shuf_lane_bytes!(bswap, 0x0c0d_0e0f_0809_0a0b, 0x0405_0607_0001_0203);
     }
 
-    impl<NI> From<x4<u128x1_sse2<YesS3, YesS4, NI>>> for u32x4x4_avx2<NI>
+    impl<NI> From<x2<u128x1_sse2<YesS3, YesS4, NI>, G0>> for u32x4x2_avx2<NI>
     where
         NI: Copy,
     {
+        #[inline(always)]
+        fn from(x: x2<u128x1_sse2<YesS3, YesS4, NI>, G0>) -> Self {
+            Self::new(unsafe { _mm256_setr_m128i(x.0[0].x, x.0[1].x) })
+        }
+    }
+
+    impl<NI> LaneWords4 for u32x4x2_avx2<NI> {
+        #[inline(always)]
+        fn shuffle_lane_words1230(self) -> Self {
+            Self::new(unsafe { _mm256_shuffle_epi32(self.x, 0b1001_0011) })
+        }
+        #[inline(always)]
+        fn shuffle_lane_words2301(self) -> Self {
+            Self::new(unsafe { _mm256_shuffle_epi32(self.x, 0b0100_1110) })
+        }
+        #[inline(always)]
+        fn shuffle_lane_words3012(self) -> Self {
+            Self::new(unsafe { _mm256_shuffle_epi32(self.x, 0b0011_1001) })
+        }
+    }
+
+    ///////////////////////////////////////////////////////////////////////////////////////////
+
+    pub type u32x4x4_avx2<NI> = x2<u32x4x2_avx2<NI>, G0>;
+    impl<NI: Copy> u32x4x4<Avx2Machine<NI>> for u32x4x4_avx2<NI> {}
+
+    impl<NI: Copy> Store<vec512_storage> for u32x4x4_avx2<NI> {
+        #[inline(always)]
+        unsafe fn unpack(p: vec512_storage) -> Self {
+            Self::new([
+                u32x4x2_avx2::unpack(p.avx[0]),
+                u32x4x2_avx2::unpack(p.avx[1]),
+            ])
+        }
+    }
+    impl<NI: Copy> MultiLane<[u32x4_sse2<YesS3, YesS4, NI>; 4]> for u32x4x4_avx2<NI> {
+        #[inline(always)]
+        fn to_lanes(self) -> [u32x4_sse2<YesS3, YesS4, NI>; 4] {
+            let [a, b] = self.0[0].to_lanes();
+            let [c, d] = self.0[1].to_lanes();
+            [a, b, c, d]
+        }
+        #[inline(always)]
+        fn from_lanes(x: [u32x4_sse2<YesS3, YesS4, NI>; 4]) -> Self {
+            let ab = u32x4x2_avx2::from_lanes([x[0], x[1]]);
+            let cd = u32x4x2_avx2::from_lanes([x[2], x[3]]);
+            Self::new([ab, cd])
+        }
+    }
+    impl<NI: Copy> Vec4<u32x4_sse2<YesS3, YesS4, NI>> for u32x4x4_avx2<NI> {
+        #[inline(always)]
+        fn extract(self, i: u32) -> u32x4_sse2<YesS3, YesS4, NI> {
+            match i {
+                0 => self.0[0].extract(0),
+                1 => self.0[0].extract(1),
+                2 => self.0[1].extract(0),
+                3 => self.0[1].extract(1),
+                _ => panic!(),
+            }
+        }
+        #[inline(always)]
+        fn insert(self, w: u32x4_sse2<YesS3, YesS4, NI>, i: u32) -> Self {
+            Self::new(match i {
+                0 | 1 => [self.0[0].insert(w, i), self.0[1]],
+                2 | 3 => [self.0[0], self.0[1].insert(w, i - 2)],
+                _ => panic!(),
+            })
+        }
+    }
+    impl<NI: Copy> Vec4Ext<u32x4_sse2<YesS3, YesS4, NI>> for u32x4x4_avx2<NI> {
+        #[inline(always)]
+        fn transpose4(a: Self, b: Self, c: Self, d: Self) -> (Self, Self, Self, Self) {
+            /*
+             * a00:a01 a10:a11
+             * b00:b01 b10:b11
+             * c00:c01 c10:c11
+             * d00:d01 d10:d11
+             *       =>
+             * a00:b00 c00:d00
+             * a01:b01 c01:d01
+             * a10:b10 c10:d10
+             * a11:b11 c11:d11
+             */
+            unsafe {
+                let ab00 = u32x4x2_avx2::new(_mm256_permute2x128_si256(a.0[0].x, b.0[0].x, 0x20));
+                let ab01 = u32x4x2_avx2::new(_mm256_permute2x128_si256(a.0[0].x, b.0[0].x, 0x31));
+                let ab10 = u32x4x2_avx2::new(_mm256_permute2x128_si256(a.0[1].x, b.0[1].x, 0x20));
+                let ab11 = u32x4x2_avx2::new(_mm256_permute2x128_si256(a.0[1].x, b.0[1].x, 0x31));
+                let cd00 = u32x4x2_avx2::new(_mm256_permute2x128_si256(c.0[0].x, d.0[0].x, 0x20));
+                let cd01 = u32x4x2_avx2::new(_mm256_permute2x128_si256(c.0[0].x, d.0[0].x, 0x31));
+                let cd10 = u32x4x2_avx2::new(_mm256_permute2x128_si256(c.0[1].x, d.0[1].x, 0x20));
+                let cd11 = u32x4x2_avx2::new(_mm256_permute2x128_si256(c.0[1].x, d.0[1].x, 0x31));
+                (
+                    Self::new([ab00, cd00]),
+                    Self::new([ab01, cd01]),
+                    Self::new([ab10, cd10]),
+                    Self::new([ab11, cd11]),
+                )
+            }
+        }
+    }
+    impl<NI: Copy> Vector<[u32; 16]> for u32x4x4_avx2<NI> {
+        #[inline(always)]
+        fn to_scalars(self) -> [u32; 16] {
+            unsafe { core::mem::transmute(self) }
+        }
+    }
+    impl<NI: Copy> From<u32x4x4_avx2<NI>> for vec512_storage {
+        #[inline(always)]
+        fn from(x: u32x4x4_avx2<NI>) -> Self {
+            Self {
+                avx: [
+                    vec256_storage { avx: x.0[0].x },
+                    vec256_storage { avx: x.0[1].x },
+                ],
+            }
+        }
+    }
+    impl<NI: Copy> From<x4<u128x1_sse2<YesS3, YesS4, NI>>> for u32x4x4_avx2<NI> {
         #[inline(always)]
         fn from(x: x4<u128x1_sse2<YesS3, YesS4, NI>>) -> Self {
             Self::new(unsafe {
                 [
-                    _mm256_setr_m128i(x.0[0].x, x.0[1].x),
-                    _mm256_setr_m128i(x.0[2].x, x.0[3].x),
+                    u32x4x2_avx2::new(_mm256_setr_m128i(x.0[0].x, x.0[1].x)),
+                    u32x4x2_avx2::new(_mm256_setr_m128i(x.0[2].x, x.0[3].x)),
                 ]
             })
         }
-- 
cgit v1.2.3