2 * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
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.
11 #include "./vpx_dsp_rtcd.h"
12 #include "vpx_dsp/x86/highbd_inv_txfm_sse2.h"
13 #include "vpx_dsp/x86/inv_txfm_sse2.h"
14 #include "vpx_dsp/x86/transpose_sse2.h"
15 #include "vpx_dsp/x86/txfm_common_sse2.h"
17 void vpx_highbd_idct4x4_16_add_sse2(const tran_low_t *input, uint16_t *dest,
19 tran_low_t out[4 * 4];
20 tran_low_t *outptr = out;
24 __m128i temp_mm, min_input, max_input;
26 int optimised_cols = 0;
27 const __m128i zero = _mm_set1_epi16(0);
28 const __m128i eight = _mm_set1_epi16(8);
29 const __m128i max = _mm_set1_epi16(12043);
30 const __m128i min = _mm_set1_epi16(-12043);
31 // Load input into __m128i
32 inptr[0] = _mm_loadu_si128((const __m128i *)input);
33 inptr[1] = _mm_loadu_si128((const __m128i *)(input + 4));
34 inptr[2] = _mm_loadu_si128((const __m128i *)(input + 8));
35 inptr[3] = _mm_loadu_si128((const __m128i *)(input + 12));
38 inptr[0] = _mm_packs_epi32(inptr[0], inptr[1]);
39 inptr[1] = _mm_packs_epi32(inptr[2], inptr[3]);
41 max_input = _mm_max_epi16(inptr[0], inptr[1]);
42 min_input = _mm_min_epi16(inptr[0], inptr[1]);
43 max_input = _mm_cmpgt_epi16(max_input, max);
44 min_input = _mm_cmplt_epi16(min_input, min);
45 temp_mm = _mm_or_si128(max_input, min_input);
46 test = _mm_movemask_epi8(temp_mm);
49 // Do the row transform
52 // Check the min & max values
53 max_input = _mm_max_epi16(inptr[0], inptr[1]);
54 min_input = _mm_min_epi16(inptr[0], inptr[1]);
55 max_input = _mm_cmpgt_epi16(max_input, max);
56 min_input = _mm_cmplt_epi16(min_input, min);
57 temp_mm = _mm_or_si128(max_input, min_input);
58 test = _mm_movemask_epi8(temp_mm);
62 sign_bits[0] = _mm_cmplt_epi16(inptr[0], zero);
63 sign_bits[1] = _mm_cmplt_epi16(inptr[1], zero);
64 inptr[3] = _mm_unpackhi_epi16(inptr[1], sign_bits[1]);
65 inptr[2] = _mm_unpacklo_epi16(inptr[1], sign_bits[1]);
66 inptr[1] = _mm_unpackhi_epi16(inptr[0], sign_bits[0]);
67 inptr[0] = _mm_unpacklo_epi16(inptr[0], sign_bits[0]);
68 _mm_storeu_si128((__m128i *)outptr, inptr[0]);
69 _mm_storeu_si128((__m128i *)(outptr + 4), inptr[1]);
70 _mm_storeu_si128((__m128i *)(outptr + 8), inptr[2]);
71 _mm_storeu_si128((__m128i *)(outptr + 12), inptr[3]);
73 // Set to use the optimised transform for the column
77 // Run the un-optimised row transform
78 for (i = 0; i < 4; ++i) {
79 vpx_highbd_idct4_c(input, outptr, bd);
88 // Final round and shift
89 inptr[0] = _mm_add_epi16(inptr[0], eight);
90 inptr[1] = _mm_add_epi16(inptr[1], eight);
92 inptr[0] = _mm_srai_epi16(inptr[0], 4);
93 inptr[1] = _mm_srai_epi16(inptr[1], 4);
95 // Reconstruction and Store
97 __m128i d0 = _mm_loadl_epi64((const __m128i *)dest);
98 __m128i d2 = _mm_loadl_epi64((const __m128i *)(dest + stride * 2));
99 d0 = _mm_unpacklo_epi64(
100 d0, _mm_loadl_epi64((const __m128i *)(dest + stride)));
101 d2 = _mm_unpacklo_epi64(
102 d2, _mm_loadl_epi64((const __m128i *)(dest + stride * 3)));
103 d0 = clamp_high_sse2(_mm_adds_epi16(d0, inptr[0]), bd);
104 d2 = clamp_high_sse2(_mm_adds_epi16(d2, inptr[1]), bd);
106 _mm_storel_epi64((__m128i *)dest, d0);
108 d0 = _mm_srli_si128(d0, 8);
109 _mm_storel_epi64((__m128i *)(dest + stride), d0);
111 _mm_storel_epi64((__m128i *)(dest + stride * 2), d2);
113 d2 = _mm_srli_si128(d2, 8);
114 _mm_storel_epi64((__m128i *)(dest + stride * 3), d2);
117 // Run the un-optimised column transform
118 tran_low_t temp_in[4], temp_out[4];
120 for (i = 0; i < 4; ++i) {
121 for (j = 0; j < 4; ++j) temp_in[j] = out[j * 4 + i];
122 vpx_highbd_idct4_c(temp_in, temp_out, bd);
123 for (j = 0; j < 4; ++j) {
124 dest[j * stride + i] = highbd_clip_pixel_add(
125 dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd);