Fix rotate_left implementation on ssse3 and avx2 for [u]int16

And also provide an optimized version for [u]int8

Author: serge-sans-paille

include/xsimd/arch/xsimd_avx2.hpp

@@ -657,11 +657,21 @@ namespace xsimd
 
         // rotate_left
         template <size_t N, class A>
-        XSIMD_INLINE batch<uint16_t, A> rotate_left(batch<uint16_t, A> const& self, requires_arch<avx2>) noexcept
+        XSIMD_INLINE batch<uint8_t, A> rotate_left(batch<uint8_t, A> const& self, requires_arch<avx2>) noexcept
         {
             return _mm256_alignr_epi8(self, self, N);
         }
         template <size_t N, class A>
+        XSIMD_INLINE batch<int8_t, A> rotate_left(batch<int8_t, A> const& self, requires_arch<avx2>) noexcept
+        {
+            return bitwise_cast<int8_t>(rotate_left<N, A>(bitwise_cast<uint8_t>(self), avx2 {}));
+        }
+        template <size_t N, class A>
+        XSIMD_INLINE batch<uint16_t, A> rotate_left(batch<uint16_t, A> const& self, requires_arch<avx2>) noexcept
+        {
+            return _mm256_alignr_epi8(self, self, 2 * N);
+        }
+        template <size_t N, class A>
         XSIMD_INLINE batch<int16_t, A> rotate_left(batch<int16_t, A> const& self, requires_arch<avx2>) noexcept
         {
             return bitwise_cast<int16_t>(rotate_left<N, A>(bitwise_cast<uint16_t>(self), avx2 {}));

include/xsimd/arch/xsimd_ssse3.hpp

@@ -107,11 +107,22 @@ namespace xsimd
 
         // rotate_left
         template <size_t N, class A>
-        XSIMD_INLINE batch<uint16_t, A> rotate_left(batch<uint16_t, A> const& self, requires_arch<ssse3>) noexcept
+        XSIMD_INLINE batch<uint8_t, A> rotate_left(batch<uint8_t, A> const& self, requires_arch<ssse3>) noexcept
         {
             return _mm_alignr_epi8(self, self, N);
         }
         template <size_t N, class A>
+        XSIMD_INLINE batch<int8_t, A> rotate_left(batch<int8_t, A> const& self, requires_arch<ssse3>) noexcept
+        {
+            return bitwise_cast<int8_t>(rotate_left<N, A>(bitwise_cast<uint8_t>(self), ssse3 {}));
+        }
+
+        template <size_t N, class A>
+        XSIMD_INLINE batch<uint16_t, A> rotate_left(batch<uint16_t, A> const& self, requires_arch<ssse3>) noexcept
+        {
+            return _mm_alignr_epi8(self, self, 2 * N);
+        }
+        template <size_t N, class A>
         XSIMD_INLINE batch<int16_t, A> rotate_left(batch<int16_t, A> const& self, requires_arch<ssse3>) noexcept
         {
             return bitwise_cast<int16_t>(rotate_left<N, A>(bitwise_cast<uint16_t>(self), ssse3 {}));