More cleanup

1 files changed, 53 insertions, 0 deletions

diff --git a/src/cube_avx2.h b/src/cube_avx2.h
index 42ca9b5..b189023 100644
--- a/src/cube_avx2.h
+++ b/src/cube_avx2.h
@@ -25,6 +25,7 @@ _static_inline cube_t compose_epcpeo(cube_t, cube_t);
 _static_inline cube_t compose_edges(cube_t, cube_t);
 _static_inline cube_t compose_corners(cube_t, cube_t);
 _static_inline cube_t compose(cube_t, cube_t);
+_static_inline cube_t inverse(cube_t);
 
 _static_inline int64_t coord_co(cube_t);
 _static_inline int64_t coord_csep(cube_t);
@@ -135,6 +136,58 @@ compose(cube_t c1, cube_t c2)
         return s;
 }
 
+_static_inline cube_t
+cleanaftershuffle(cube_t c)
+{
+        __m256i b;
+
+        b = _mm256_set_epi8(
+                ~0, ~0, ~0, ~0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                ~0, ~0, ~0, ~0, ~0, ~0, ~0, ~0, 0, 0, 0, 0, 0, 0, 0, 0
+        );
+
+        return _mm256_andnot_si256(b, c);
+}
+
+_static_inline cube_t
+inverse(cube_t c)
+{
+        /* Method taken from Andrew Skalski's vcube[1]. The addition sequence
+         * was generated using [2].
+         * [1] https://github.com/Voltara/vcube
+         * [2] http://wwwhomes.uni-bielefeld.de/achim/addition_chain.html
+         */
+        cube_t v3, vi, vo, vp, ret;
+
+        v3 = _mm256_shuffle_epi8(c, c);
+        v3 = _mm256_shuffle_epi8(v3, c);
+        vi = _mm256_shuffle_epi8(v3, v3);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, v3);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, c);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, v3);
+        vi = _mm256_shuffle_epi8(vi, vi);
+        vi = _mm256_shuffle_epi8(vi, c);
+
+        vo = _mm256_and_si256(c, _mm256_or_si256(_eo_avx2, _co2_avx2));
+        vo = _mm256_shuffle_epi8(vo, vi);
+        vp = _mm256_andnot_si256(_mm256_or_si256(_eo_avx2, _co2_avx2), vi);
+        ret = _mm256_or_si256(vp, vo);
+        ret = cleanaftershuffle(ret);
+        
+        return invertco(ret);
+}
+
 _static_inline int64_t
 coord_co(cube_t c)
 {