1/*
2 *  Copyright (c) 2015 The WebM project authors. All Rights Reserved.
3 *
4 *  Use of this source code is governed by a BSD-style license
5 *  that can be found in the LICENSE file in the root of the source
6 *  tree. An additional intellectual property rights grant can be found
7 *  in the file PATENTS.  All contributing project authors may
8 *  be found in the AUTHORS file in the root of the source tree.
9 */
10
11#ifndef VPX_DSP_X86_TRANSPOSE_SSE2_H_
12#define VPX_DSP_X86_TRANSPOSE_SSE2_H_
13
14#include "./vpx_dsp_rtcd.h"
15#include "vpx_dsp/x86/inv_txfm_sse2.h"
16#include "vpx_dsp/x86/txfm_common_sse2.h"
17
18static INLINE void transpose_16bit_4x4(__m128i *res) {
19  const __m128i tr0_0 = _mm_unpacklo_epi16(res[0], res[1]);
20  const __m128i tr0_1 = _mm_unpackhi_epi16(res[0], res[1]);
21
22  res[0] = _mm_unpacklo_epi16(tr0_0, tr0_1);
23  res[1] = _mm_unpackhi_epi16(tr0_0, tr0_1);
24}
25
26static INLINE void transpose_32bit_4x4(__m128i *const a0, __m128i *const a1,
27                                       __m128i *const a2, __m128i *const a3) {
28  // Unpack 32 bit elements. Goes from:
29  // a0: 00 01 02 03
30  // a1: 10 11 12 13
31  // a2: 20 21 22 23
32  // a3: 30 31 32 33
33  // to:
34  // b0: 00 10 01 11
35  // b1: 20 30 21 31
36  // b2: 02 12 03 13
37  // b3: 22 32 23 33
38
39  const __m128i b0 = _mm_unpacklo_epi32(*a0, *a1);
40  const __m128i b1 = _mm_unpacklo_epi32(*a2, *a3);
41  const __m128i b2 = _mm_unpackhi_epi32(*a0, *a1);
42  const __m128i b3 = _mm_unpackhi_epi32(*a2, *a3);
43
44  // Unpack 64 bit elements resulting in:
45  // a0: 00 10 20 30
46  // a1: 01 11 21 31
47  // a2: 02 12 22 32
48  // a3: 03 13 23 33
49  *a0 = _mm_unpacklo_epi64(b0, b1);
50  *a1 = _mm_unpackhi_epi64(b0, b1);
51  *a2 = _mm_unpacklo_epi64(b2, b3);
52  *a3 = _mm_unpackhi_epi64(b2, b3);
53}
54
55#endif  // VPX_DSP_X86_TRANSPOSE_SSE2_H_
56