123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652 |
- #ifndef OPENCV_CORE_SSE_UTILS_HPP
- #define OPENCV_CORE_SSE_UTILS_HPP
- #ifndef __cplusplus
- # error sse_utils.hpp header must be compiled as C++
- #endif
- #include "opencv2/core/cvdef.h"
- #if CV_SSE2
- inline void _mm_deinterleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1)
- {
- __m128i layer1_chunk0 = _mm_unpacklo_epi8(v_r0, v_g0);
- __m128i layer1_chunk1 = _mm_unpackhi_epi8(v_r0, v_g0);
- __m128i layer1_chunk2 = _mm_unpacklo_epi8(v_r1, v_g1);
- __m128i layer1_chunk3 = _mm_unpackhi_epi8(v_r1, v_g1);
- __m128i layer2_chunk0 = _mm_unpacklo_epi8(layer1_chunk0, layer1_chunk2);
- __m128i layer2_chunk1 = _mm_unpackhi_epi8(layer1_chunk0, layer1_chunk2);
- __m128i layer2_chunk2 = _mm_unpacklo_epi8(layer1_chunk1, layer1_chunk3);
- __m128i layer2_chunk3 = _mm_unpackhi_epi8(layer1_chunk1, layer1_chunk3);
- __m128i layer3_chunk0 = _mm_unpacklo_epi8(layer2_chunk0, layer2_chunk2);
- __m128i layer3_chunk1 = _mm_unpackhi_epi8(layer2_chunk0, layer2_chunk2);
- __m128i layer3_chunk2 = _mm_unpacklo_epi8(layer2_chunk1, layer2_chunk3);
- __m128i layer3_chunk3 = _mm_unpackhi_epi8(layer2_chunk1, layer2_chunk3);
- __m128i layer4_chunk0 = _mm_unpacklo_epi8(layer3_chunk0, layer3_chunk2);
- __m128i layer4_chunk1 = _mm_unpackhi_epi8(layer3_chunk0, layer3_chunk2);
- __m128i layer4_chunk2 = _mm_unpacklo_epi8(layer3_chunk1, layer3_chunk3);
- __m128i layer4_chunk3 = _mm_unpackhi_epi8(layer3_chunk1, layer3_chunk3);
- v_r0 = _mm_unpacklo_epi8(layer4_chunk0, layer4_chunk2);
- v_r1 = _mm_unpackhi_epi8(layer4_chunk0, layer4_chunk2);
- v_g0 = _mm_unpacklo_epi8(layer4_chunk1, layer4_chunk3);
- v_g1 = _mm_unpackhi_epi8(layer4_chunk1, layer4_chunk3);
- }
- inline void _mm_deinterleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
- __m128i & v_g1, __m128i & v_b0, __m128i & v_b1)
- {
- __m128i layer1_chunk0 = _mm_unpacklo_epi8(v_r0, v_g1);
- __m128i layer1_chunk1 = _mm_unpackhi_epi8(v_r0, v_g1);
- __m128i layer1_chunk2 = _mm_unpacklo_epi8(v_r1, v_b0);
- __m128i layer1_chunk3 = _mm_unpackhi_epi8(v_r1, v_b0);
- __m128i layer1_chunk4 = _mm_unpacklo_epi8(v_g0, v_b1);
- __m128i layer1_chunk5 = _mm_unpackhi_epi8(v_g0, v_b1);
- __m128i layer2_chunk0 = _mm_unpacklo_epi8(layer1_chunk0, layer1_chunk3);
- __m128i layer2_chunk1 = _mm_unpackhi_epi8(layer1_chunk0, layer1_chunk3);
- __m128i layer2_chunk2 = _mm_unpacklo_epi8(layer1_chunk1, layer1_chunk4);
- __m128i layer2_chunk3 = _mm_unpackhi_epi8(layer1_chunk1, layer1_chunk4);
- __m128i layer2_chunk4 = _mm_unpacklo_epi8(layer1_chunk2, layer1_chunk5);
- __m128i layer2_chunk5 = _mm_unpackhi_epi8(layer1_chunk2, layer1_chunk5);
- __m128i layer3_chunk0 = _mm_unpacklo_epi8(layer2_chunk0, layer2_chunk3);
- __m128i layer3_chunk1 = _mm_unpackhi_epi8(layer2_chunk0, layer2_chunk3);
- __m128i layer3_chunk2 = _mm_unpacklo_epi8(layer2_chunk1, layer2_chunk4);
- __m128i layer3_chunk3 = _mm_unpackhi_epi8(layer2_chunk1, layer2_chunk4);
- __m128i layer3_chunk4 = _mm_unpacklo_epi8(layer2_chunk2, layer2_chunk5);
- __m128i layer3_chunk5 = _mm_unpackhi_epi8(layer2_chunk2, layer2_chunk5);
- __m128i layer4_chunk0 = _mm_unpacklo_epi8(layer3_chunk0, layer3_chunk3);
- __m128i layer4_chunk1 = _mm_unpackhi_epi8(layer3_chunk0, layer3_chunk3);
- __m128i layer4_chunk2 = _mm_unpacklo_epi8(layer3_chunk1, layer3_chunk4);
- __m128i layer4_chunk3 = _mm_unpackhi_epi8(layer3_chunk1, layer3_chunk4);
- __m128i layer4_chunk4 = _mm_unpacklo_epi8(layer3_chunk2, layer3_chunk5);
- __m128i layer4_chunk5 = _mm_unpackhi_epi8(layer3_chunk2, layer3_chunk5);
- v_r0 = _mm_unpacklo_epi8(layer4_chunk0, layer4_chunk3);
- v_r1 = _mm_unpackhi_epi8(layer4_chunk0, layer4_chunk3);
- v_g0 = _mm_unpacklo_epi8(layer4_chunk1, layer4_chunk4);
- v_g1 = _mm_unpackhi_epi8(layer4_chunk1, layer4_chunk4);
- v_b0 = _mm_unpacklo_epi8(layer4_chunk2, layer4_chunk5);
- v_b1 = _mm_unpackhi_epi8(layer4_chunk2, layer4_chunk5);
- }
- inline void _mm_deinterleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1,
- __m128i & v_b0, __m128i & v_b1, __m128i & v_a0, __m128i & v_a1)
- {
- __m128i layer1_chunk0 = _mm_unpacklo_epi8(v_r0, v_b0);
- __m128i layer1_chunk1 = _mm_unpackhi_epi8(v_r0, v_b0);
- __m128i layer1_chunk2 = _mm_unpacklo_epi8(v_r1, v_b1);
- __m128i layer1_chunk3 = _mm_unpackhi_epi8(v_r1, v_b1);
- __m128i layer1_chunk4 = _mm_unpacklo_epi8(v_g0, v_a0);
- __m128i layer1_chunk5 = _mm_unpackhi_epi8(v_g0, v_a0);
- __m128i layer1_chunk6 = _mm_unpacklo_epi8(v_g1, v_a1);
- __m128i layer1_chunk7 = _mm_unpackhi_epi8(v_g1, v_a1);
- __m128i layer2_chunk0 = _mm_unpacklo_epi8(layer1_chunk0, layer1_chunk4);
- __m128i layer2_chunk1 = _mm_unpackhi_epi8(layer1_chunk0, layer1_chunk4);
- __m128i layer2_chunk2 = _mm_unpacklo_epi8(layer1_chunk1, layer1_chunk5);
- __m128i layer2_chunk3 = _mm_unpackhi_epi8(layer1_chunk1, layer1_chunk5);
- __m128i layer2_chunk4 = _mm_unpacklo_epi8(layer1_chunk2, layer1_chunk6);
- __m128i layer2_chunk5 = _mm_unpackhi_epi8(layer1_chunk2, layer1_chunk6);
- __m128i layer2_chunk6 = _mm_unpacklo_epi8(layer1_chunk3, layer1_chunk7);
- __m128i layer2_chunk7 = _mm_unpackhi_epi8(layer1_chunk3, layer1_chunk7);
- __m128i layer3_chunk0 = _mm_unpacklo_epi8(layer2_chunk0, layer2_chunk4);
- __m128i layer3_chunk1 = _mm_unpackhi_epi8(layer2_chunk0, layer2_chunk4);
- __m128i layer3_chunk2 = _mm_unpacklo_epi8(layer2_chunk1, layer2_chunk5);
- __m128i layer3_chunk3 = _mm_unpackhi_epi8(layer2_chunk1, layer2_chunk5);
- __m128i layer3_chunk4 = _mm_unpacklo_epi8(layer2_chunk2, layer2_chunk6);
- __m128i layer3_chunk5 = _mm_unpackhi_epi8(layer2_chunk2, layer2_chunk6);
- __m128i layer3_chunk6 = _mm_unpacklo_epi8(layer2_chunk3, layer2_chunk7);
- __m128i layer3_chunk7 = _mm_unpackhi_epi8(layer2_chunk3, layer2_chunk7);
- __m128i layer4_chunk0 = _mm_unpacklo_epi8(layer3_chunk0, layer3_chunk4);
- __m128i layer4_chunk1 = _mm_unpackhi_epi8(layer3_chunk0, layer3_chunk4);
- __m128i layer4_chunk2 = _mm_unpacklo_epi8(layer3_chunk1, layer3_chunk5);
- __m128i layer4_chunk3 = _mm_unpackhi_epi8(layer3_chunk1, layer3_chunk5);
- __m128i layer4_chunk4 = _mm_unpacklo_epi8(layer3_chunk2, layer3_chunk6);
- __m128i layer4_chunk5 = _mm_unpackhi_epi8(layer3_chunk2, layer3_chunk6);
- __m128i layer4_chunk6 = _mm_unpacklo_epi8(layer3_chunk3, layer3_chunk7);
- __m128i layer4_chunk7 = _mm_unpackhi_epi8(layer3_chunk3, layer3_chunk7);
- v_r0 = _mm_unpacklo_epi8(layer4_chunk0, layer4_chunk4);
- v_r1 = _mm_unpackhi_epi8(layer4_chunk0, layer4_chunk4);
- v_g0 = _mm_unpacklo_epi8(layer4_chunk1, layer4_chunk5);
- v_g1 = _mm_unpackhi_epi8(layer4_chunk1, layer4_chunk5);
- v_b0 = _mm_unpacklo_epi8(layer4_chunk2, layer4_chunk6);
- v_b1 = _mm_unpackhi_epi8(layer4_chunk2, layer4_chunk6);
- v_a0 = _mm_unpacklo_epi8(layer4_chunk3, layer4_chunk7);
- v_a1 = _mm_unpackhi_epi8(layer4_chunk3, layer4_chunk7);
- }
- inline void _mm_interleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1)
- {
- __m128i v_mask = _mm_set1_epi16(0x00ff);
- __m128i layer4_chunk0 = _mm_packus_epi16(_mm_and_si128(v_r0, v_mask), _mm_and_si128(v_r1, v_mask));
- __m128i layer4_chunk2 = _mm_packus_epi16(_mm_srli_epi16(v_r0, 8), _mm_srli_epi16(v_r1, 8));
- __m128i layer4_chunk1 = _mm_packus_epi16(_mm_and_si128(v_g0, v_mask), _mm_and_si128(v_g1, v_mask));
- __m128i layer4_chunk3 = _mm_packus_epi16(_mm_srli_epi16(v_g0, 8), _mm_srli_epi16(v_g1, 8));
- __m128i layer3_chunk0 = _mm_packus_epi16(_mm_and_si128(layer4_chunk0, v_mask), _mm_and_si128(layer4_chunk1, v_mask));
- __m128i layer3_chunk2 = _mm_packus_epi16(_mm_srli_epi16(layer4_chunk0, 8), _mm_srli_epi16(layer4_chunk1, 8));
- __m128i layer3_chunk1 = _mm_packus_epi16(_mm_and_si128(layer4_chunk2, v_mask), _mm_and_si128(layer4_chunk3, v_mask));
- __m128i layer3_chunk3 = _mm_packus_epi16(_mm_srli_epi16(layer4_chunk2, 8), _mm_srli_epi16(layer4_chunk3, 8));
- __m128i layer2_chunk0 = _mm_packus_epi16(_mm_and_si128(layer3_chunk0, v_mask), _mm_and_si128(layer3_chunk1, v_mask));
- __m128i layer2_chunk2 = _mm_packus_epi16(_mm_srli_epi16(layer3_chunk0, 8), _mm_srli_epi16(layer3_chunk1, 8));
- __m128i layer2_chunk1 = _mm_packus_epi16(_mm_and_si128(layer3_chunk2, v_mask), _mm_and_si128(layer3_chunk3, v_mask));
- __m128i layer2_chunk3 = _mm_packus_epi16(_mm_srli_epi16(layer3_chunk2, 8), _mm_srli_epi16(layer3_chunk3, 8));
- __m128i layer1_chunk0 = _mm_packus_epi16(_mm_and_si128(layer2_chunk0, v_mask), _mm_and_si128(layer2_chunk1, v_mask));
- __m128i layer1_chunk2 = _mm_packus_epi16(_mm_srli_epi16(layer2_chunk0, 8), _mm_srli_epi16(layer2_chunk1, 8));
- __m128i layer1_chunk1 = _mm_packus_epi16(_mm_and_si128(layer2_chunk2, v_mask), _mm_and_si128(layer2_chunk3, v_mask));
- __m128i layer1_chunk3 = _mm_packus_epi16(_mm_srli_epi16(layer2_chunk2, 8), _mm_srli_epi16(layer2_chunk3, 8));
- v_r0 = _mm_packus_epi16(_mm_and_si128(layer1_chunk0, v_mask), _mm_and_si128(layer1_chunk1, v_mask));
- v_g0 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk0, 8), _mm_srli_epi16(layer1_chunk1, 8));
- v_r1 = _mm_packus_epi16(_mm_and_si128(layer1_chunk2, v_mask), _mm_and_si128(layer1_chunk3, v_mask));
- v_g1 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk2, 8), _mm_srli_epi16(layer1_chunk3, 8));
- }
- inline void _mm_interleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
- __m128i & v_g1, __m128i & v_b0, __m128i & v_b1)
- {
- __m128i v_mask = _mm_set1_epi16(0x00ff);
- __m128i layer4_chunk0 = _mm_packus_epi16(_mm_and_si128(v_r0, v_mask), _mm_and_si128(v_r1, v_mask));
- __m128i layer4_chunk3 = _mm_packus_epi16(_mm_srli_epi16(v_r0, 8), _mm_srli_epi16(v_r1, 8));
- __m128i layer4_chunk1 = _mm_packus_epi16(_mm_and_si128(v_g0, v_mask), _mm_and_si128(v_g1, v_mask));
- __m128i layer4_chunk4 = _mm_packus_epi16(_mm_srli_epi16(v_g0, 8), _mm_srli_epi16(v_g1, 8));
- __m128i layer4_chunk2 = _mm_packus_epi16(_mm_and_si128(v_b0, v_mask), _mm_and_si128(v_b1, v_mask));
- __m128i layer4_chunk5 = _mm_packus_epi16(_mm_srli_epi16(v_b0, 8), _mm_srli_epi16(v_b1, 8));
- __m128i layer3_chunk0 = _mm_packus_epi16(_mm_and_si128(layer4_chunk0, v_mask), _mm_and_si128(layer4_chunk1, v_mask));
- __m128i layer3_chunk3 = _mm_packus_epi16(_mm_srli_epi16(layer4_chunk0, 8), _mm_srli_epi16(layer4_chunk1, 8));
- __m128i layer3_chunk1 = _mm_packus_epi16(_mm_and_si128(layer4_chunk2, v_mask), _mm_and_si128(layer4_chunk3, v_mask));
- __m128i layer3_chunk4 = _mm_packus_epi16(_mm_srli_epi16(layer4_chunk2, 8), _mm_srli_epi16(layer4_chunk3, 8));
- __m128i layer3_chunk2 = _mm_packus_epi16(_mm_and_si128(layer4_chunk4, v_mask), _mm_and_si128(layer4_chunk5, v_mask));
- __m128i layer3_chunk5 = _mm_packus_epi16(_mm_srli_epi16(layer4_chunk4, 8), _mm_srli_epi16(layer4_chunk5, 8));
- __m128i layer2_chunk0 = _mm_packus_epi16(_mm_and_si128(layer3_chunk0, v_mask), _mm_and_si128(layer3_chunk1, v_mask));
- __m128i layer2_chunk3 = _mm_packus_epi16(_mm_srli_epi16(layer3_chunk0, 8), _mm_srli_epi16(layer3_chunk1, 8));
- __m128i layer2_chunk1 = _mm_packus_epi16(_mm_and_si128(layer3_chunk2, v_mask), _mm_and_si128(layer3_chunk3, v_mask));
- __m128i layer2_chunk4 = _mm_packus_epi16(_mm_srli_epi16(layer3_chunk2, 8), _mm_srli_epi16(layer3_chunk3, 8));
- __m128i layer2_chunk2 = _mm_packus_epi16(_mm_and_si128(layer3_chunk4, v_mask), _mm_and_si128(layer3_chunk5, v_mask));
- __m128i layer2_chunk5 = _mm_packus_epi16(_mm_srli_epi16(layer3_chunk4, 8), _mm_srli_epi16(layer3_chunk5, 8));
- __m128i layer1_chunk0 = _mm_packus_epi16(_mm_and_si128(layer2_chunk0, v_mask), _mm_and_si128(layer2_chunk1, v_mask));
- __m128i layer1_chunk3 = _mm_packus_epi16(_mm_srli_epi16(layer2_chunk0, 8), _mm_srli_epi16(layer2_chunk1, 8));
- __m128i layer1_chunk1 = _mm_packus_epi16(_mm_and_si128(layer2_chunk2, v_mask), _mm_and_si128(layer2_chunk3, v_mask));
- __m128i layer1_chunk4 = _mm_packus_epi16(_mm_srli_epi16(layer2_chunk2, 8), _mm_srli_epi16(layer2_chunk3, 8));
- __m128i layer1_chunk2 = _mm_packus_epi16(_mm_and_si128(layer2_chunk4, v_mask), _mm_and_si128(layer2_chunk5, v_mask));
- __m128i layer1_chunk5 = _mm_packus_epi16(_mm_srli_epi16(layer2_chunk4, 8), _mm_srli_epi16(layer2_chunk5, 8));
- v_r0 = _mm_packus_epi16(_mm_and_si128(layer1_chunk0, v_mask), _mm_and_si128(layer1_chunk1, v_mask));
- v_g1 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk0, 8), _mm_srli_epi16(layer1_chunk1, 8));
- v_r1 = _mm_packus_epi16(_mm_and_si128(layer1_chunk2, v_mask), _mm_and_si128(layer1_chunk3, v_mask));
- v_b0 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk2, 8), _mm_srli_epi16(layer1_chunk3, 8));
- v_g0 = _mm_packus_epi16(_mm_and_si128(layer1_chunk4, v_mask), _mm_and_si128(layer1_chunk5, v_mask));
- v_b1 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk4, 8), _mm_srli_epi16(layer1_chunk5, 8));
- }
- inline void _mm_interleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1,
- __m128i & v_b0, __m128i & v_b1, __m128i & v_a0, __m128i & v_a1)
- {
- __m128i v_mask = _mm_set1_epi16(0x00ff);
- __m128i layer4_chunk0 = _mm_packus_epi16(_mm_and_si128(v_r0, v_mask), _mm_and_si128(v_r1, v_mask));
- __m128i layer4_chunk4 = _mm_packus_epi16(_mm_srli_epi16(v_r0, 8), _mm_srli_epi16(v_r1, 8));
- __m128i layer4_chunk1 = _mm_packus_epi16(_mm_and_si128(v_g0, v_mask), _mm_and_si128(v_g1, v_mask));
- __m128i layer4_chunk5 = _mm_packus_epi16(_mm_srli_epi16(v_g0, 8), _mm_srli_epi16(v_g1, 8));
- __m128i layer4_chunk2 = _mm_packus_epi16(_mm_and_si128(v_b0, v_mask), _mm_and_si128(v_b1, v_mask));
- __m128i layer4_chunk6 = _mm_packus_epi16(_mm_srli_epi16(v_b0, 8), _mm_srli_epi16(v_b1, 8));
- __m128i layer4_chunk3 = _mm_packus_epi16(_mm_and_si128(v_a0, v_mask), _mm_and_si128(v_a1, v_mask));
- __m128i layer4_chunk7 = _mm_packus_epi16(_mm_srli_epi16(v_a0, 8), _mm_srli_epi16(v_a1, 8));
- __m128i layer3_chunk0 = _mm_packus_epi16(_mm_and_si128(layer4_chunk0, v_mask), _mm_and_si128(layer4_chunk1, v_mask));
- __m128i layer3_chunk4 = _mm_packus_epi16(_mm_srli_epi16(layer4_chunk0, 8), _mm_srli_epi16(layer4_chunk1, 8));
- __m128i layer3_chunk1 = _mm_packus_epi16(_mm_and_si128(layer4_chunk2, v_mask), _mm_and_si128(layer4_chunk3, v_mask));
- __m128i layer3_chunk5 = _mm_packus_epi16(_mm_srli_epi16(layer4_chunk2, 8), _mm_srli_epi16(layer4_chunk3, 8));
- __m128i layer3_chunk2 = _mm_packus_epi16(_mm_and_si128(layer4_chunk4, v_mask), _mm_and_si128(layer4_chunk5, v_mask));
- __m128i layer3_chunk6 = _mm_packus_epi16(_mm_srli_epi16(layer4_chunk4, 8), _mm_srli_epi16(layer4_chunk5, 8));
- __m128i layer3_chunk3 = _mm_packus_epi16(_mm_and_si128(layer4_chunk6, v_mask), _mm_and_si128(layer4_chunk7, v_mask));
- __m128i layer3_chunk7 = _mm_packus_epi16(_mm_srli_epi16(layer4_chunk6, 8), _mm_srli_epi16(layer4_chunk7, 8));
- __m128i layer2_chunk0 = _mm_packus_epi16(_mm_and_si128(layer3_chunk0, v_mask), _mm_and_si128(layer3_chunk1, v_mask));
- __m128i layer2_chunk4 = _mm_packus_epi16(_mm_srli_epi16(layer3_chunk0, 8), _mm_srli_epi16(layer3_chunk1, 8));
- __m128i layer2_chunk1 = _mm_packus_epi16(_mm_and_si128(layer3_chunk2, v_mask), _mm_and_si128(layer3_chunk3, v_mask));
- __m128i layer2_chunk5 = _mm_packus_epi16(_mm_srli_epi16(layer3_chunk2, 8), _mm_srli_epi16(layer3_chunk3, 8));
- __m128i layer2_chunk2 = _mm_packus_epi16(_mm_and_si128(layer3_chunk4, v_mask), _mm_and_si128(layer3_chunk5, v_mask));
- __m128i layer2_chunk6 = _mm_packus_epi16(_mm_srli_epi16(layer3_chunk4, 8), _mm_srli_epi16(layer3_chunk5, 8));
- __m128i layer2_chunk3 = _mm_packus_epi16(_mm_and_si128(layer3_chunk6, v_mask), _mm_and_si128(layer3_chunk7, v_mask));
- __m128i layer2_chunk7 = _mm_packus_epi16(_mm_srli_epi16(layer3_chunk6, 8), _mm_srli_epi16(layer3_chunk7, 8));
- __m128i layer1_chunk0 = _mm_packus_epi16(_mm_and_si128(layer2_chunk0, v_mask), _mm_and_si128(layer2_chunk1, v_mask));
- __m128i layer1_chunk4 = _mm_packus_epi16(_mm_srli_epi16(layer2_chunk0, 8), _mm_srli_epi16(layer2_chunk1, 8));
- __m128i layer1_chunk1 = _mm_packus_epi16(_mm_and_si128(layer2_chunk2, v_mask), _mm_and_si128(layer2_chunk3, v_mask));
- __m128i layer1_chunk5 = _mm_packus_epi16(_mm_srli_epi16(layer2_chunk2, 8), _mm_srli_epi16(layer2_chunk3, 8));
- __m128i layer1_chunk2 = _mm_packus_epi16(_mm_and_si128(layer2_chunk4, v_mask), _mm_and_si128(layer2_chunk5, v_mask));
- __m128i layer1_chunk6 = _mm_packus_epi16(_mm_srli_epi16(layer2_chunk4, 8), _mm_srli_epi16(layer2_chunk5, 8));
- __m128i layer1_chunk3 = _mm_packus_epi16(_mm_and_si128(layer2_chunk6, v_mask), _mm_and_si128(layer2_chunk7, v_mask));
- __m128i layer1_chunk7 = _mm_packus_epi16(_mm_srli_epi16(layer2_chunk6, 8), _mm_srli_epi16(layer2_chunk7, 8));
- v_r0 = _mm_packus_epi16(_mm_and_si128(layer1_chunk0, v_mask), _mm_and_si128(layer1_chunk1, v_mask));
- v_b0 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk0, 8), _mm_srli_epi16(layer1_chunk1, 8));
- v_r1 = _mm_packus_epi16(_mm_and_si128(layer1_chunk2, v_mask), _mm_and_si128(layer1_chunk3, v_mask));
- v_b1 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk2, 8), _mm_srli_epi16(layer1_chunk3, 8));
- v_g0 = _mm_packus_epi16(_mm_and_si128(layer1_chunk4, v_mask), _mm_and_si128(layer1_chunk5, v_mask));
- v_a0 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk4, 8), _mm_srli_epi16(layer1_chunk5, 8));
- v_g1 = _mm_packus_epi16(_mm_and_si128(layer1_chunk6, v_mask), _mm_and_si128(layer1_chunk7, v_mask));
- v_a1 = _mm_packus_epi16(_mm_srli_epi16(layer1_chunk6, 8), _mm_srli_epi16(layer1_chunk7, 8));
- }
- inline void _mm_deinterleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1)
- {
- __m128i layer1_chunk0 = _mm_unpacklo_epi16(v_r0, v_g0);
- __m128i layer1_chunk1 = _mm_unpackhi_epi16(v_r0, v_g0);
- __m128i layer1_chunk2 = _mm_unpacklo_epi16(v_r1, v_g1);
- __m128i layer1_chunk3 = _mm_unpackhi_epi16(v_r1, v_g1);
- __m128i layer2_chunk0 = _mm_unpacklo_epi16(layer1_chunk0, layer1_chunk2);
- __m128i layer2_chunk1 = _mm_unpackhi_epi16(layer1_chunk0, layer1_chunk2);
- __m128i layer2_chunk2 = _mm_unpacklo_epi16(layer1_chunk1, layer1_chunk3);
- __m128i layer2_chunk3 = _mm_unpackhi_epi16(layer1_chunk1, layer1_chunk3);
- __m128i layer3_chunk0 = _mm_unpacklo_epi16(layer2_chunk0, layer2_chunk2);
- __m128i layer3_chunk1 = _mm_unpackhi_epi16(layer2_chunk0, layer2_chunk2);
- __m128i layer3_chunk2 = _mm_unpacklo_epi16(layer2_chunk1, layer2_chunk3);
- __m128i layer3_chunk3 = _mm_unpackhi_epi16(layer2_chunk1, layer2_chunk3);
- v_r0 = _mm_unpacklo_epi16(layer3_chunk0, layer3_chunk2);
- v_r1 = _mm_unpackhi_epi16(layer3_chunk0, layer3_chunk2);
- v_g0 = _mm_unpacklo_epi16(layer3_chunk1, layer3_chunk3);
- v_g1 = _mm_unpackhi_epi16(layer3_chunk1, layer3_chunk3);
- }
- inline void _mm_deinterleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
- __m128i & v_g1, __m128i & v_b0, __m128i & v_b1)
- {
- __m128i layer1_chunk0 = _mm_unpacklo_epi16(v_r0, v_g1);
- __m128i layer1_chunk1 = _mm_unpackhi_epi16(v_r0, v_g1);
- __m128i layer1_chunk2 = _mm_unpacklo_epi16(v_r1, v_b0);
- __m128i layer1_chunk3 = _mm_unpackhi_epi16(v_r1, v_b0);
- __m128i layer1_chunk4 = _mm_unpacklo_epi16(v_g0, v_b1);
- __m128i layer1_chunk5 = _mm_unpackhi_epi16(v_g0, v_b1);
- __m128i layer2_chunk0 = _mm_unpacklo_epi16(layer1_chunk0, layer1_chunk3);
- __m128i layer2_chunk1 = _mm_unpackhi_epi16(layer1_chunk0, layer1_chunk3);
- __m128i layer2_chunk2 = _mm_unpacklo_epi16(layer1_chunk1, layer1_chunk4);
- __m128i layer2_chunk3 = _mm_unpackhi_epi16(layer1_chunk1, layer1_chunk4);
- __m128i layer2_chunk4 = _mm_unpacklo_epi16(layer1_chunk2, layer1_chunk5);
- __m128i layer2_chunk5 = _mm_unpackhi_epi16(layer1_chunk2, layer1_chunk5);
- __m128i layer3_chunk0 = _mm_unpacklo_epi16(layer2_chunk0, layer2_chunk3);
- __m128i layer3_chunk1 = _mm_unpackhi_epi16(layer2_chunk0, layer2_chunk3);
- __m128i layer3_chunk2 = _mm_unpacklo_epi16(layer2_chunk1, layer2_chunk4);
- __m128i layer3_chunk3 = _mm_unpackhi_epi16(layer2_chunk1, layer2_chunk4);
- __m128i layer3_chunk4 = _mm_unpacklo_epi16(layer2_chunk2, layer2_chunk5);
- __m128i layer3_chunk5 = _mm_unpackhi_epi16(layer2_chunk2, layer2_chunk5);
- v_r0 = _mm_unpacklo_epi16(layer3_chunk0, layer3_chunk3);
- v_r1 = _mm_unpackhi_epi16(layer3_chunk0, layer3_chunk3);
- v_g0 = _mm_unpacklo_epi16(layer3_chunk1, layer3_chunk4);
- v_g1 = _mm_unpackhi_epi16(layer3_chunk1, layer3_chunk4);
- v_b0 = _mm_unpacklo_epi16(layer3_chunk2, layer3_chunk5);
- v_b1 = _mm_unpackhi_epi16(layer3_chunk2, layer3_chunk5);
- }
- inline void _mm_deinterleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1,
- __m128i & v_b0, __m128i & v_b1, __m128i & v_a0, __m128i & v_a1)
- {
- __m128i layer1_chunk0 = _mm_unpacklo_epi16(v_r0, v_b0);
- __m128i layer1_chunk1 = _mm_unpackhi_epi16(v_r0, v_b0);
- __m128i layer1_chunk2 = _mm_unpacklo_epi16(v_r1, v_b1);
- __m128i layer1_chunk3 = _mm_unpackhi_epi16(v_r1, v_b1);
- __m128i layer1_chunk4 = _mm_unpacklo_epi16(v_g0, v_a0);
- __m128i layer1_chunk5 = _mm_unpackhi_epi16(v_g0, v_a0);
- __m128i layer1_chunk6 = _mm_unpacklo_epi16(v_g1, v_a1);
- __m128i layer1_chunk7 = _mm_unpackhi_epi16(v_g1, v_a1);
- __m128i layer2_chunk0 = _mm_unpacklo_epi16(layer1_chunk0, layer1_chunk4);
- __m128i layer2_chunk1 = _mm_unpackhi_epi16(layer1_chunk0, layer1_chunk4);
- __m128i layer2_chunk2 = _mm_unpacklo_epi16(layer1_chunk1, layer1_chunk5);
- __m128i layer2_chunk3 = _mm_unpackhi_epi16(layer1_chunk1, layer1_chunk5);
- __m128i layer2_chunk4 = _mm_unpacklo_epi16(layer1_chunk2, layer1_chunk6);
- __m128i layer2_chunk5 = _mm_unpackhi_epi16(layer1_chunk2, layer1_chunk6);
- __m128i layer2_chunk6 = _mm_unpacklo_epi16(layer1_chunk3, layer1_chunk7);
- __m128i layer2_chunk7 = _mm_unpackhi_epi16(layer1_chunk3, layer1_chunk7);
- __m128i layer3_chunk0 = _mm_unpacklo_epi16(layer2_chunk0, layer2_chunk4);
- __m128i layer3_chunk1 = _mm_unpackhi_epi16(layer2_chunk0, layer2_chunk4);
- __m128i layer3_chunk2 = _mm_unpacklo_epi16(layer2_chunk1, layer2_chunk5);
- __m128i layer3_chunk3 = _mm_unpackhi_epi16(layer2_chunk1, layer2_chunk5);
- __m128i layer3_chunk4 = _mm_unpacklo_epi16(layer2_chunk2, layer2_chunk6);
- __m128i layer3_chunk5 = _mm_unpackhi_epi16(layer2_chunk2, layer2_chunk6);
- __m128i layer3_chunk6 = _mm_unpacklo_epi16(layer2_chunk3, layer2_chunk7);
- __m128i layer3_chunk7 = _mm_unpackhi_epi16(layer2_chunk3, layer2_chunk7);
- v_r0 = _mm_unpacklo_epi16(layer3_chunk0, layer3_chunk4);
- v_r1 = _mm_unpackhi_epi16(layer3_chunk0, layer3_chunk4);
- v_g0 = _mm_unpacklo_epi16(layer3_chunk1, layer3_chunk5);
- v_g1 = _mm_unpackhi_epi16(layer3_chunk1, layer3_chunk5);
- v_b0 = _mm_unpacklo_epi16(layer3_chunk2, layer3_chunk6);
- v_b1 = _mm_unpackhi_epi16(layer3_chunk2, layer3_chunk6);
- v_a0 = _mm_unpacklo_epi16(layer3_chunk3, layer3_chunk7);
- v_a1 = _mm_unpackhi_epi16(layer3_chunk3, layer3_chunk7);
- }
- #if CV_SSE4_1
- inline void _mm_interleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1)
- {
- __m128i v_mask = _mm_set1_epi32(0x0000ffff);
- __m128i layer3_chunk0 = _mm_packus_epi32(_mm_and_si128(v_r0, v_mask), _mm_and_si128(v_r1, v_mask));
- __m128i layer3_chunk2 = _mm_packus_epi32(_mm_srli_epi32(v_r0, 16), _mm_srli_epi32(v_r1, 16));
- __m128i layer3_chunk1 = _mm_packus_epi32(_mm_and_si128(v_g0, v_mask), _mm_and_si128(v_g1, v_mask));
- __m128i layer3_chunk3 = _mm_packus_epi32(_mm_srli_epi32(v_g0, 16), _mm_srli_epi32(v_g1, 16));
- __m128i layer2_chunk0 = _mm_packus_epi32(_mm_and_si128(layer3_chunk0, v_mask), _mm_and_si128(layer3_chunk1, v_mask));
- __m128i layer2_chunk2 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk0, 16), _mm_srli_epi32(layer3_chunk1, 16));
- __m128i layer2_chunk1 = _mm_packus_epi32(_mm_and_si128(layer3_chunk2, v_mask), _mm_and_si128(layer3_chunk3, v_mask));
- __m128i layer2_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk2, 16), _mm_srli_epi32(layer3_chunk3, 16));
- __m128i layer1_chunk0 = _mm_packus_epi32(_mm_and_si128(layer2_chunk0, v_mask), _mm_and_si128(layer2_chunk1, v_mask));
- __m128i layer1_chunk2 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk0, 16), _mm_srli_epi32(layer2_chunk1, 16));
- __m128i layer1_chunk1 = _mm_packus_epi32(_mm_and_si128(layer2_chunk2, v_mask), _mm_and_si128(layer2_chunk3, v_mask));
- __m128i layer1_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk2, 16), _mm_srli_epi32(layer2_chunk3, 16));
- v_r0 = _mm_packus_epi32(_mm_and_si128(layer1_chunk0, v_mask), _mm_and_si128(layer1_chunk1, v_mask));
- v_g0 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk0, 16), _mm_srli_epi32(layer1_chunk1, 16));
- v_r1 = _mm_packus_epi32(_mm_and_si128(layer1_chunk2, v_mask), _mm_and_si128(layer1_chunk3, v_mask));
- v_g1 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk2, 16), _mm_srli_epi32(layer1_chunk3, 16));
- }
- inline void _mm_interleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0,
- __m128i & v_g1, __m128i & v_b0, __m128i & v_b1)
- {
- __m128i v_mask = _mm_set1_epi32(0x0000ffff);
- __m128i layer3_chunk0 = _mm_packus_epi32(_mm_and_si128(v_r0, v_mask), _mm_and_si128(v_r1, v_mask));
- __m128i layer3_chunk3 = _mm_packus_epi32(_mm_srli_epi32(v_r0, 16), _mm_srli_epi32(v_r1, 16));
- __m128i layer3_chunk1 = _mm_packus_epi32(_mm_and_si128(v_g0, v_mask), _mm_and_si128(v_g1, v_mask));
- __m128i layer3_chunk4 = _mm_packus_epi32(_mm_srli_epi32(v_g0, 16), _mm_srli_epi32(v_g1, 16));
- __m128i layer3_chunk2 = _mm_packus_epi32(_mm_and_si128(v_b0, v_mask), _mm_and_si128(v_b1, v_mask));
- __m128i layer3_chunk5 = _mm_packus_epi32(_mm_srli_epi32(v_b0, 16), _mm_srli_epi32(v_b1, 16));
- __m128i layer2_chunk0 = _mm_packus_epi32(_mm_and_si128(layer3_chunk0, v_mask), _mm_and_si128(layer3_chunk1, v_mask));
- __m128i layer2_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk0, 16), _mm_srli_epi32(layer3_chunk1, 16));
- __m128i layer2_chunk1 = _mm_packus_epi32(_mm_and_si128(layer3_chunk2, v_mask), _mm_and_si128(layer3_chunk3, v_mask));
- __m128i layer2_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk2, 16), _mm_srli_epi32(layer3_chunk3, 16));
- __m128i layer2_chunk2 = _mm_packus_epi32(_mm_and_si128(layer3_chunk4, v_mask), _mm_and_si128(layer3_chunk5, v_mask));
- __m128i layer2_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk4, 16), _mm_srli_epi32(layer3_chunk5, 16));
- __m128i layer1_chunk0 = _mm_packus_epi32(_mm_and_si128(layer2_chunk0, v_mask), _mm_and_si128(layer2_chunk1, v_mask));
- __m128i layer1_chunk3 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk0, 16), _mm_srli_epi32(layer2_chunk1, 16));
- __m128i layer1_chunk1 = _mm_packus_epi32(_mm_and_si128(layer2_chunk2, v_mask), _mm_and_si128(layer2_chunk3, v_mask));
- __m128i layer1_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk2, 16), _mm_srli_epi32(layer2_chunk3, 16));
- __m128i layer1_chunk2 = _mm_packus_epi32(_mm_and_si128(layer2_chunk4, v_mask), _mm_and_si128(layer2_chunk5, v_mask));
- __m128i layer1_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk4, 16), _mm_srli_epi32(layer2_chunk5, 16));
- v_r0 = _mm_packus_epi32(_mm_and_si128(layer1_chunk0, v_mask), _mm_and_si128(layer1_chunk1, v_mask));
- v_g1 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk0, 16), _mm_srli_epi32(layer1_chunk1, 16));
- v_r1 = _mm_packus_epi32(_mm_and_si128(layer1_chunk2, v_mask), _mm_and_si128(layer1_chunk3, v_mask));
- v_b0 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk2, 16), _mm_srli_epi32(layer1_chunk3, 16));
- v_g0 = _mm_packus_epi32(_mm_and_si128(layer1_chunk4, v_mask), _mm_and_si128(layer1_chunk5, v_mask));
- v_b1 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk4, 16), _mm_srli_epi32(layer1_chunk5, 16));
- }
- inline void _mm_interleave_epi16(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1,
- __m128i & v_b0, __m128i & v_b1, __m128i & v_a0, __m128i & v_a1)
- {
- __m128i v_mask = _mm_set1_epi32(0x0000ffff);
- __m128i layer3_chunk0 = _mm_packus_epi32(_mm_and_si128(v_r0, v_mask), _mm_and_si128(v_r1, v_mask));
- __m128i layer3_chunk4 = _mm_packus_epi32(_mm_srli_epi32(v_r0, 16), _mm_srli_epi32(v_r1, 16));
- __m128i layer3_chunk1 = _mm_packus_epi32(_mm_and_si128(v_g0, v_mask), _mm_and_si128(v_g1, v_mask));
- __m128i layer3_chunk5 = _mm_packus_epi32(_mm_srli_epi32(v_g0, 16), _mm_srli_epi32(v_g1, 16));
- __m128i layer3_chunk2 = _mm_packus_epi32(_mm_and_si128(v_b0, v_mask), _mm_and_si128(v_b1, v_mask));
- __m128i layer3_chunk6 = _mm_packus_epi32(_mm_srli_epi32(v_b0, 16), _mm_srli_epi32(v_b1, 16));
- __m128i layer3_chunk3 = _mm_packus_epi32(_mm_and_si128(v_a0, v_mask), _mm_and_si128(v_a1, v_mask));
- __m128i layer3_chunk7 = _mm_packus_epi32(_mm_srli_epi32(v_a0, 16), _mm_srli_epi32(v_a1, 16));
- __m128i layer2_chunk0 = _mm_packus_epi32(_mm_and_si128(layer3_chunk0, v_mask), _mm_and_si128(layer3_chunk1, v_mask));
- __m128i layer2_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk0, 16), _mm_srli_epi32(layer3_chunk1, 16));
- __m128i layer2_chunk1 = _mm_packus_epi32(_mm_and_si128(layer3_chunk2, v_mask), _mm_and_si128(layer3_chunk3, v_mask));
- __m128i layer2_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk2, 16), _mm_srli_epi32(layer3_chunk3, 16));
- __m128i layer2_chunk2 = _mm_packus_epi32(_mm_and_si128(layer3_chunk4, v_mask), _mm_and_si128(layer3_chunk5, v_mask));
- __m128i layer2_chunk6 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk4, 16), _mm_srli_epi32(layer3_chunk5, 16));
- __m128i layer2_chunk3 = _mm_packus_epi32(_mm_and_si128(layer3_chunk6, v_mask), _mm_and_si128(layer3_chunk7, v_mask));
- __m128i layer2_chunk7 = _mm_packus_epi32(_mm_srli_epi32(layer3_chunk6, 16), _mm_srli_epi32(layer3_chunk7, 16));
- __m128i layer1_chunk0 = _mm_packus_epi32(_mm_and_si128(layer2_chunk0, v_mask), _mm_and_si128(layer2_chunk1, v_mask));
- __m128i layer1_chunk4 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk0, 16), _mm_srli_epi32(layer2_chunk1, 16));
- __m128i layer1_chunk1 = _mm_packus_epi32(_mm_and_si128(layer2_chunk2, v_mask), _mm_and_si128(layer2_chunk3, v_mask));
- __m128i layer1_chunk5 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk2, 16), _mm_srli_epi32(layer2_chunk3, 16));
- __m128i layer1_chunk2 = _mm_packus_epi32(_mm_and_si128(layer2_chunk4, v_mask), _mm_and_si128(layer2_chunk5, v_mask));
- __m128i layer1_chunk6 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk4, 16), _mm_srli_epi32(layer2_chunk5, 16));
- __m128i layer1_chunk3 = _mm_packus_epi32(_mm_and_si128(layer2_chunk6, v_mask), _mm_and_si128(layer2_chunk7, v_mask));
- __m128i layer1_chunk7 = _mm_packus_epi32(_mm_srli_epi32(layer2_chunk6, 16), _mm_srli_epi32(layer2_chunk7, 16));
- v_r0 = _mm_packus_epi32(_mm_and_si128(layer1_chunk0, v_mask), _mm_and_si128(layer1_chunk1, v_mask));
- v_b0 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk0, 16), _mm_srli_epi32(layer1_chunk1, 16));
- v_r1 = _mm_packus_epi32(_mm_and_si128(layer1_chunk2, v_mask), _mm_and_si128(layer1_chunk3, v_mask));
- v_b1 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk2, 16), _mm_srli_epi32(layer1_chunk3, 16));
- v_g0 = _mm_packus_epi32(_mm_and_si128(layer1_chunk4, v_mask), _mm_and_si128(layer1_chunk5, v_mask));
- v_a0 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk4, 16), _mm_srli_epi32(layer1_chunk5, 16));
- v_g1 = _mm_packus_epi32(_mm_and_si128(layer1_chunk6, v_mask), _mm_and_si128(layer1_chunk7, v_mask));
- v_a1 = _mm_packus_epi32(_mm_srli_epi32(layer1_chunk6, 16), _mm_srli_epi32(layer1_chunk7, 16));
- }
- #endif
- inline void _mm_deinterleave_ps(__m128 & v_r0, __m128 & v_r1, __m128 & v_g0, __m128 & v_g1)
- {
- __m128 layer1_chunk0 = _mm_unpacklo_ps(v_r0, v_g0);
- __m128 layer1_chunk1 = _mm_unpackhi_ps(v_r0, v_g0);
- __m128 layer1_chunk2 = _mm_unpacklo_ps(v_r1, v_g1);
- __m128 layer1_chunk3 = _mm_unpackhi_ps(v_r1, v_g1);
- __m128 layer2_chunk0 = _mm_unpacklo_ps(layer1_chunk0, layer1_chunk2);
- __m128 layer2_chunk1 = _mm_unpackhi_ps(layer1_chunk0, layer1_chunk2);
- __m128 layer2_chunk2 = _mm_unpacklo_ps(layer1_chunk1, layer1_chunk3);
- __m128 layer2_chunk3 = _mm_unpackhi_ps(layer1_chunk1, layer1_chunk3);
- v_r0 = _mm_unpacklo_ps(layer2_chunk0, layer2_chunk2);
- v_r1 = _mm_unpackhi_ps(layer2_chunk0, layer2_chunk2);
- v_g0 = _mm_unpacklo_ps(layer2_chunk1, layer2_chunk3);
- v_g1 = _mm_unpackhi_ps(layer2_chunk1, layer2_chunk3);
- }
- inline void _mm_deinterleave_ps(__m128 & v_r0, __m128 & v_r1, __m128 & v_g0,
- __m128 & v_g1, __m128 & v_b0, __m128 & v_b1)
- {
- __m128 layer1_chunk0 = _mm_unpacklo_ps(v_r0, v_g1);
- __m128 layer1_chunk1 = _mm_unpackhi_ps(v_r0, v_g1);
- __m128 layer1_chunk2 = _mm_unpacklo_ps(v_r1, v_b0);
- __m128 layer1_chunk3 = _mm_unpackhi_ps(v_r1, v_b0);
- __m128 layer1_chunk4 = _mm_unpacklo_ps(v_g0, v_b1);
- __m128 layer1_chunk5 = _mm_unpackhi_ps(v_g0, v_b1);
- __m128 layer2_chunk0 = _mm_unpacklo_ps(layer1_chunk0, layer1_chunk3);
- __m128 layer2_chunk1 = _mm_unpackhi_ps(layer1_chunk0, layer1_chunk3);
- __m128 layer2_chunk2 = _mm_unpacklo_ps(layer1_chunk1, layer1_chunk4);
- __m128 layer2_chunk3 = _mm_unpackhi_ps(layer1_chunk1, layer1_chunk4);
- __m128 layer2_chunk4 = _mm_unpacklo_ps(layer1_chunk2, layer1_chunk5);
- __m128 layer2_chunk5 = _mm_unpackhi_ps(layer1_chunk2, layer1_chunk5);
- v_r0 = _mm_unpacklo_ps(layer2_chunk0, layer2_chunk3);
- v_r1 = _mm_unpackhi_ps(layer2_chunk0, layer2_chunk3);
- v_g0 = _mm_unpacklo_ps(layer2_chunk1, layer2_chunk4);
- v_g1 = _mm_unpackhi_ps(layer2_chunk1, layer2_chunk4);
- v_b0 = _mm_unpacklo_ps(layer2_chunk2, layer2_chunk5);
- v_b1 = _mm_unpackhi_ps(layer2_chunk2, layer2_chunk5);
- }
- inline void _mm_deinterleave_ps(__m128 & v_r0, __m128 & v_r1, __m128 & v_g0, __m128 & v_g1,
- __m128 & v_b0, __m128 & v_b1, __m128 & v_a0, __m128 & v_a1)
- {
- __m128 layer1_chunk0 = _mm_unpacklo_ps(v_r0, v_b0);
- __m128 layer1_chunk1 = _mm_unpackhi_ps(v_r0, v_b0);
- __m128 layer1_chunk2 = _mm_unpacklo_ps(v_r1, v_b1);
- __m128 layer1_chunk3 = _mm_unpackhi_ps(v_r1, v_b1);
- __m128 layer1_chunk4 = _mm_unpacklo_ps(v_g0, v_a0);
- __m128 layer1_chunk5 = _mm_unpackhi_ps(v_g0, v_a0);
- __m128 layer1_chunk6 = _mm_unpacklo_ps(v_g1, v_a1);
- __m128 layer1_chunk7 = _mm_unpackhi_ps(v_g1, v_a1);
- __m128 layer2_chunk0 = _mm_unpacklo_ps(layer1_chunk0, layer1_chunk4);
- __m128 layer2_chunk1 = _mm_unpackhi_ps(layer1_chunk0, layer1_chunk4);
- __m128 layer2_chunk2 = _mm_unpacklo_ps(layer1_chunk1, layer1_chunk5);
- __m128 layer2_chunk3 = _mm_unpackhi_ps(layer1_chunk1, layer1_chunk5);
- __m128 layer2_chunk4 = _mm_unpacklo_ps(layer1_chunk2, layer1_chunk6);
- __m128 layer2_chunk5 = _mm_unpackhi_ps(layer1_chunk2, layer1_chunk6);
- __m128 layer2_chunk6 = _mm_unpacklo_ps(layer1_chunk3, layer1_chunk7);
- __m128 layer2_chunk7 = _mm_unpackhi_ps(layer1_chunk3, layer1_chunk7);
- v_r0 = _mm_unpacklo_ps(layer2_chunk0, layer2_chunk4);
- v_r1 = _mm_unpackhi_ps(layer2_chunk0, layer2_chunk4);
- v_g0 = _mm_unpacklo_ps(layer2_chunk1, layer2_chunk5);
- v_g1 = _mm_unpackhi_ps(layer2_chunk1, layer2_chunk5);
- v_b0 = _mm_unpacklo_ps(layer2_chunk2, layer2_chunk6);
- v_b1 = _mm_unpackhi_ps(layer2_chunk2, layer2_chunk6);
- v_a0 = _mm_unpacklo_ps(layer2_chunk3, layer2_chunk7);
- v_a1 = _mm_unpackhi_ps(layer2_chunk3, layer2_chunk7);
- }
- inline void _mm_interleave_ps(__m128 & v_r0, __m128 & v_r1, __m128 & v_g0, __m128 & v_g1)
- {
- const int mask_lo = _MM_SHUFFLE(2, 0, 2, 0), mask_hi = _MM_SHUFFLE(3, 1, 3, 1);
- __m128 layer2_chunk0 = _mm_shuffle_ps(v_r0, v_r1, mask_lo);
- __m128 layer2_chunk2 = _mm_shuffle_ps(v_r0, v_r1, mask_hi);
- __m128 layer2_chunk1 = _mm_shuffle_ps(v_g0, v_g1, mask_lo);
- __m128 layer2_chunk3 = _mm_shuffle_ps(v_g0, v_g1, mask_hi);
- __m128 layer1_chunk0 = _mm_shuffle_ps(layer2_chunk0, layer2_chunk1, mask_lo);
- __m128 layer1_chunk2 = _mm_shuffle_ps(layer2_chunk0, layer2_chunk1, mask_hi);
- __m128 layer1_chunk1 = _mm_shuffle_ps(layer2_chunk2, layer2_chunk3, mask_lo);
- __m128 layer1_chunk3 = _mm_shuffle_ps(layer2_chunk2, layer2_chunk3, mask_hi);
- v_r0 = _mm_shuffle_ps(layer1_chunk0, layer1_chunk1, mask_lo);
- v_g0 = _mm_shuffle_ps(layer1_chunk0, layer1_chunk1, mask_hi);
- v_r1 = _mm_shuffle_ps(layer1_chunk2, layer1_chunk3, mask_lo);
- v_g1 = _mm_shuffle_ps(layer1_chunk2, layer1_chunk3, mask_hi);
- }
- inline void _mm_interleave_ps(__m128 & v_r0, __m128 & v_r1, __m128 & v_g0,
- __m128 & v_g1, __m128 & v_b0, __m128 & v_b1)
- {
- const int mask_lo = _MM_SHUFFLE(2, 0, 2, 0), mask_hi = _MM_SHUFFLE(3, 1, 3, 1);
- __m128 layer2_chunk0 = _mm_shuffle_ps(v_r0, v_r1, mask_lo);
- __m128 layer2_chunk3 = _mm_shuffle_ps(v_r0, v_r1, mask_hi);
- __m128 layer2_chunk1 = _mm_shuffle_ps(v_g0, v_g1, mask_lo);
- __m128 layer2_chunk4 = _mm_shuffle_ps(v_g0, v_g1, mask_hi);
- __m128 layer2_chunk2 = _mm_shuffle_ps(v_b0, v_b1, mask_lo);
- __m128 layer2_chunk5 = _mm_shuffle_ps(v_b0, v_b1, mask_hi);
- __m128 layer1_chunk0 = _mm_shuffle_ps(layer2_chunk0, layer2_chunk1, mask_lo);
- __m128 layer1_chunk3 = _mm_shuffle_ps(layer2_chunk0, layer2_chunk1, mask_hi);
- __m128 layer1_chunk1 = _mm_shuffle_ps(layer2_chunk2, layer2_chunk3, mask_lo);
- __m128 layer1_chunk4 = _mm_shuffle_ps(layer2_chunk2, layer2_chunk3, mask_hi);
- __m128 layer1_chunk2 = _mm_shuffle_ps(layer2_chunk4, layer2_chunk5, mask_lo);
- __m128 layer1_chunk5 = _mm_shuffle_ps(layer2_chunk4, layer2_chunk5, mask_hi);
- v_r0 = _mm_shuffle_ps(layer1_chunk0, layer1_chunk1, mask_lo);
- v_g1 = _mm_shuffle_ps(layer1_chunk0, layer1_chunk1, mask_hi);
- v_r1 = _mm_shuffle_ps(layer1_chunk2, layer1_chunk3, mask_lo);
- v_b0 = _mm_shuffle_ps(layer1_chunk2, layer1_chunk3, mask_hi);
- v_g0 = _mm_shuffle_ps(layer1_chunk4, layer1_chunk5, mask_lo);
- v_b1 = _mm_shuffle_ps(layer1_chunk4, layer1_chunk5, mask_hi);
- }
- inline void _mm_interleave_ps(__m128 & v_r0, __m128 & v_r1, __m128 & v_g0, __m128 & v_g1,
- __m128 & v_b0, __m128 & v_b1, __m128 & v_a0, __m128 & v_a1)
- {
- const int mask_lo = _MM_SHUFFLE(2, 0, 2, 0), mask_hi = _MM_SHUFFLE(3, 1, 3, 1);
- __m128 layer2_chunk0 = _mm_shuffle_ps(v_r0, v_r1, mask_lo);
- __m128 layer2_chunk4 = _mm_shuffle_ps(v_r0, v_r1, mask_hi);
- __m128 layer2_chunk1 = _mm_shuffle_ps(v_g0, v_g1, mask_lo);
- __m128 layer2_chunk5 = _mm_shuffle_ps(v_g0, v_g1, mask_hi);
- __m128 layer2_chunk2 = _mm_shuffle_ps(v_b0, v_b1, mask_lo);
- __m128 layer2_chunk6 = _mm_shuffle_ps(v_b0, v_b1, mask_hi);
- __m128 layer2_chunk3 = _mm_shuffle_ps(v_a0, v_a1, mask_lo);
- __m128 layer2_chunk7 = _mm_shuffle_ps(v_a0, v_a1, mask_hi);
- __m128 layer1_chunk0 = _mm_shuffle_ps(layer2_chunk0, layer2_chunk1, mask_lo);
- __m128 layer1_chunk4 = _mm_shuffle_ps(layer2_chunk0, layer2_chunk1, mask_hi);
- __m128 layer1_chunk1 = _mm_shuffle_ps(layer2_chunk2, layer2_chunk3, mask_lo);
- __m128 layer1_chunk5 = _mm_shuffle_ps(layer2_chunk2, layer2_chunk3, mask_hi);
- __m128 layer1_chunk2 = _mm_shuffle_ps(layer2_chunk4, layer2_chunk5, mask_lo);
- __m128 layer1_chunk6 = _mm_shuffle_ps(layer2_chunk4, layer2_chunk5, mask_hi);
- __m128 layer1_chunk3 = _mm_shuffle_ps(layer2_chunk6, layer2_chunk7, mask_lo);
- __m128 layer1_chunk7 = _mm_shuffle_ps(layer2_chunk6, layer2_chunk7, mask_hi);
- v_r0 = _mm_shuffle_ps(layer1_chunk0, layer1_chunk1, mask_lo);
- v_b0 = _mm_shuffle_ps(layer1_chunk0, layer1_chunk1, mask_hi);
- v_r1 = _mm_shuffle_ps(layer1_chunk2, layer1_chunk3, mask_lo);
- v_b1 = _mm_shuffle_ps(layer1_chunk2, layer1_chunk3, mask_hi);
- v_g0 = _mm_shuffle_ps(layer1_chunk4, layer1_chunk5, mask_lo);
- v_a0 = _mm_shuffle_ps(layer1_chunk4, layer1_chunk5, mask_hi);
- v_g1 = _mm_shuffle_ps(layer1_chunk6, layer1_chunk7, mask_lo);
- v_a1 = _mm_shuffle_ps(layer1_chunk6, layer1_chunk7, mask_hi);
- }
- #endif
- #endif
|