Add slide_left&slide_right for avx512vbmi

Author: junparser

include/xsimd/arch/xsimd_avx512vbmi.hpp

@@ -17,4 +17,66 @@
 
 #include "../types/xsimd_avx512vbmi_register.hpp"
 
+namespace xsimd
+{
+
+    namespace kernel
+    {
+        using namespace types;
+
+        namespace detail
+        {
+            template <size_t N, size_t... Is>
+            constexpr std::array<uint8_t, sizeof...(Is)> make_slide_left_bytes_pattern(::xsimd::detail::index_sequence<Is...>)
+            {
+                return { (Is >= N ? Is - N : 0)... };
+            }
+
+            template <size_t N, size_t... Is>
+            constexpr std::array<uint8_t, sizeof...(Is)> make_slide_right_bytes_pattern(::xsimd::detail::index_sequence<Is...>)
+            {
+                return { (Is < (64 - N) ? Is + N : 0)... };
+            }
+        }
+
+        // slide_left
+        template <size_t N, class A, class T>
+        XSIMD_INLINE batch<T, A> slide_left(batch<T, A> const& x, requires_arch<avx512vbmi>) noexcept
+        {
+            constexpr unsigned BitCount = N * 8;
+            if (BitCount == 0)
+            {
+                return x;
+            }
+            if (BitCount >= 512)
+            {
+                return batch<T, A>(T(0));
+            }
+
+            __mmask64 mask = 0xFFFFFFFFFFFFFFFFull << N;
+            alignas(A::alignment()) auto slide_pattern = detail::make_slide_left_bytes_pattern<N>(::xsimd::detail::make_index_sequence<512 / 8>());
+            return _mm512_maskz_permutexvar_epi8(mask, _mm512_load_epi32(slide_pattern.data()), x);
+        }
+
+        // slide_right
+        template <size_t N, class A, class T>
+        XSIMD_INLINE batch<T, A> slide_right(batch<T, A> const& x, requires_arch<avx512vbmi>) noexcept
+        {
+            constexpr unsigned BitCount = N * 8;
+            if (BitCount == 0)
+            {
+                return x;
+            }
+            if (BitCount >= 512)
+            {
+                return batch<T, A>(T(0));
+            }
+            __mmask64 mask = 0xFFFFFFFFFFFFFFFFull >> N;
+            alignas(A::alignment()) auto slide_pattern = detail::make_slide_right_bytes_pattern<N>(::xsimd::detail::make_index_sequence<512 / 8>());
+            return _mm512_maskz_permutexvar_epi8(mask, _mm512_load_epi32(slide_pattern.data()), x);
+        }
+
+    }
+}
+
 #endif