Coverage Report

Created: 2023-06-07 06:31

/src/aom/aom_dsp/x86/synonyms_avx2.h
Line
Count
Source (jump to first uncovered line)
1
/*
2
 * Copyright (c) 2018, Alliance for Open Media. All rights reserved
3
 *
4
 * This source code is subject to the terms of the BSD 2 Clause License and
5
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6
 * was not distributed with this source code in the LICENSE file, you can
7
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8
 * Media Patent License 1.0 was not distributed with this source code in the
9
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10
 */
11
12
#ifndef AOM_AOM_DSP_X86_SYNONYMS_AVX2_H_
13
#define AOM_AOM_DSP_X86_SYNONYMS_AVX2_H_
14
15
#include <immintrin.h>
16
17
#include "config/aom_config.h"
18
19
#include "aom/aom_integer.h"
20
21
/**
22
 * Various reusable shorthands for x86 SIMD intrinsics.
23
 *
24
 * Intrinsics prefixed with xx_ operate on or return 128bit XMM registers.
25
 * Intrinsics prefixed with yy_ operate on or return 256bit YMM registers.
26
 */
27
28
// Loads and stores to do away with the tedium of casting the address
29
// to the right type.
30
367M
static INLINE __m256i yy_load_256(const void *a) {
31
367M
  return _mm256_load_si256((const __m256i *)a);
32
367M
}
Unexecuted instantiation: blend_a64_mask_avx2.c:yy_load_256
Unexecuted instantiation: jnt_convolve_avx2.c:yy_load_256
Unexecuted instantiation: reconinter_avx2.c:yy_load_256
selfguided_avx2.c:yy_load_256
Line
Count
Source
30
367M
static INLINE __m256i yy_load_256(const void *a) {
31
367M
  return _mm256_load_si256((const __m256i *)a);
32
367M
}
Unexecuted instantiation: wiener_convolve_avx2.c:yy_load_256
Unexecuted instantiation: highbd_wiener_convolve_avx2.c:yy_load_256
33
34
2.70G
static INLINE __m256i yy_loadu_256(const void *a) {
35
2.70G
  return _mm256_loadu_si256((const __m256i *)a);
36
2.70G
}
blend_a64_mask_avx2.c:yy_loadu_256
Line
Count
Source
34
27.1M
static INLINE __m256i yy_loadu_256(const void *a) {
35
27.1M
  return _mm256_loadu_si256((const __m256i *)a);
36
27.1M
}
Unexecuted instantiation: jnt_convolve_avx2.c:yy_loadu_256
reconinter_avx2.c:yy_loadu_256
Line
Count
Source
34
11.7M
static INLINE __m256i yy_loadu_256(const void *a) {
35
11.7M
  return _mm256_loadu_si256((const __m256i *)a);
36
11.7M
}
selfguided_avx2.c:yy_loadu_256
Line
Count
Source
34
2.50G
static INLINE __m256i yy_loadu_256(const void *a) {
35
2.50G
  return _mm256_loadu_si256((const __m256i *)a);
36
2.50G
}
Unexecuted instantiation: wiener_convolve_avx2.c:yy_loadu_256
highbd_wiener_convolve_avx2.c:yy_loadu_256
Line
Count
Source
34
168M
static INLINE __m256i yy_loadu_256(const void *a) {
35
168M
  return _mm256_loadu_si256((const __m256i *)a);
36
168M
}
37
38
366M
static INLINE void yy_store_256(void *const a, const __m256i v) {
39
366M
  _mm256_store_si256((__m256i *)a, v);
40
366M
}
Unexecuted instantiation: blend_a64_mask_avx2.c:yy_store_256
Unexecuted instantiation: jnt_convolve_avx2.c:yy_store_256
Unexecuted instantiation: reconinter_avx2.c:yy_store_256
selfguided_avx2.c:yy_store_256
Line
Count
Source
38
366M
static INLINE void yy_store_256(void *const a, const __m256i v) {
39
366M
  _mm256_store_si256((__m256i *)a, v);
40
366M
}
Unexecuted instantiation: wiener_convolve_avx2.c:yy_store_256
Unexecuted instantiation: highbd_wiener_convolve_avx2.c:yy_store_256
41
42
413M
static INLINE void yy_storeu_256(void *const a, const __m256i v) {
43
413M
  _mm256_storeu_si256((__m256i *)a, v);
44
413M
}
blend_a64_mask_avx2.c:yy_storeu_256
Line
Count
Source
42
3.00M
static INLINE void yy_storeu_256(void *const a, const __m256i v) {
43
3.00M
  _mm256_storeu_si256((__m256i *)a, v);
44
3.00M
}
Unexecuted instantiation: jnt_convolve_avx2.c:yy_storeu_256
reconinter_avx2.c:yy_storeu_256
Line
Count
Source
42
3.06M
static INLINE void yy_storeu_256(void *const a, const __m256i v) {
43
3.06M
  _mm256_storeu_si256((__m256i *)a, v);
44
3.06M
}
selfguided_avx2.c:yy_storeu_256
Line
Count
Source
42
386M
static INLINE void yy_storeu_256(void *const a, const __m256i v) {
43
386M
  _mm256_storeu_si256((__m256i *)a, v);
44
386M
}
Unexecuted instantiation: wiener_convolve_avx2.c:yy_storeu_256
highbd_wiener_convolve_avx2.c:yy_storeu_256
Line
Count
Source
42
21.0M
static INLINE void yy_storeu_256(void *const a, const __m256i v) {
43
21.0M
  _mm256_storeu_si256((__m256i *)a, v);
44
21.0M
}
45
46
// The _mm256_set1_epi64x() intrinsic is undefined for some Visual Studio
47
// compilers. The following function is equivalent to _mm256_set1_epi64x()
48
// acting on a 32-bit integer.
49
0
static INLINE __m256i yy_set1_64_from_32i(int32_t a) {
50
0
#if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900
51
0
  return _mm256_set_epi32(0, a, 0, a, 0, a, 0, a);
52
0
#else
53
0
  return _mm256_set1_epi64x((uint32_t)a);
54
0
#endif
55
0
}
Unexecuted instantiation: blend_a64_mask_avx2.c:yy_set1_64_from_32i
Unexecuted instantiation: jnt_convolve_avx2.c:yy_set1_64_from_32i
Unexecuted instantiation: reconinter_avx2.c:yy_set1_64_from_32i
Unexecuted instantiation: selfguided_avx2.c:yy_set1_64_from_32i
Unexecuted instantiation: wiener_convolve_avx2.c:yy_set1_64_from_32i
Unexecuted instantiation: highbd_wiener_convolve_avx2.c:yy_set1_64_from_32i
56
57
// Some compilers don't have _mm256_set_m128i defined in immintrin.h. We
58
// therefore define an equivalent function using a different intrinsic.
59
// ([ hi ], [ lo ]) -> [ hi ][ lo ]
60
2.15M
static INLINE __m256i yy_set_m128i(__m128i hi, __m128i lo) {
61
2.15M
  return _mm256_insertf128_si256(_mm256_castsi128_si256(lo), hi, 1);
62
2.15M
}
blend_a64_mask_avx2.c:yy_set_m128i
Line
Count
Source
60
776k
static INLINE __m256i yy_set_m128i(__m128i hi, __m128i lo) {
61
776k
  return _mm256_insertf128_si256(_mm256_castsi128_si256(lo), hi, 1);
62
776k
}
Unexecuted instantiation: jnt_convolve_avx2.c:yy_set_m128i
reconinter_avx2.c:yy_set_m128i
Line
Count
Source
60
457k
static INLINE __m256i yy_set_m128i(__m128i hi, __m128i lo) {
61
457k
  return _mm256_insertf128_si256(_mm256_castsi128_si256(lo), hi, 1);
62
457k
}
Unexecuted instantiation: selfguided_avx2.c:yy_set_m128i
Unexecuted instantiation: wiener_convolve_avx2.c:yy_set_m128i
highbd_wiener_convolve_avx2.c:yy_set_m128i
Line
Count
Source
60
921k
static INLINE __m256i yy_set_m128i(__m128i hi, __m128i lo) {
61
921k
  return _mm256_insertf128_si256(_mm256_castsi128_si256(lo), hi, 1);
62
921k
}
63
64
1.23M
static INLINE __m256i yy_loadu2_128(const void *hi, const void *lo) {
65
1.23M
  __m128i mhi = _mm_loadu_si128((__m128i *)(hi));
66
1.23M
  __m128i mlo = _mm_loadu_si128((__m128i *)(lo));
67
1.23M
  return yy_set_m128i(mhi, mlo);
68
1.23M
}
blend_a64_mask_avx2.c:yy_loadu2_128
Line
Count
Source
64
776k
static INLINE __m256i yy_loadu2_128(const void *hi, const void *lo) {
65
776k
  __m128i mhi = _mm_loadu_si128((__m128i *)(hi));
66
776k
  __m128i mlo = _mm_loadu_si128((__m128i *)(lo));
67
776k
  return yy_set_m128i(mhi, mlo);
68
776k
}
Unexecuted instantiation: jnt_convolve_avx2.c:yy_loadu2_128
reconinter_avx2.c:yy_loadu2_128
Line
Count
Source
64
457k
static INLINE __m256i yy_loadu2_128(const void *hi, const void *lo) {
65
457k
  __m128i mhi = _mm_loadu_si128((__m128i *)(hi));
66
457k
  __m128i mlo = _mm_loadu_si128((__m128i *)(lo));
67
457k
  return yy_set_m128i(mhi, mlo);
68
457k
}
Unexecuted instantiation: selfguided_avx2.c:yy_loadu2_128
Unexecuted instantiation: wiener_convolve_avx2.c:yy_loadu2_128
Unexecuted instantiation: highbd_wiener_convolve_avx2.c:yy_loadu2_128
69
70
265k
static INLINE void yy_storeu2_128(void *hi, void *lo, const __m256i a) {
71
265k
  _mm_storeu_si128((__m128i *)hi, _mm256_extracti128_si256(a, 1));
72
265k
  _mm_storeu_si128((__m128i *)lo, _mm256_castsi256_si128(a));
73
265k
}
blend_a64_mask_avx2.c:yy_storeu2_128
Line
Count
Source
70
265k
static INLINE void yy_storeu2_128(void *hi, void *lo, const __m256i a) {
71
265k
  _mm_storeu_si128((__m128i *)hi, _mm256_extracti128_si256(a, 1));
72
265k
  _mm_storeu_si128((__m128i *)lo, _mm256_castsi256_si128(a));
73
265k
}
Unexecuted instantiation: jnt_convolve_avx2.c:yy_storeu2_128
Unexecuted instantiation: reconinter_avx2.c:yy_storeu2_128
Unexecuted instantiation: selfguided_avx2.c:yy_storeu2_128
Unexecuted instantiation: wiener_convolve_avx2.c:yy_storeu2_128
Unexecuted instantiation: highbd_wiener_convolve_avx2.c:yy_storeu2_128
74
75
1.37M
static INLINE __m256i yy_roundn_epu16(__m256i v_val_w, int bits) {
76
1.37M
  const __m256i v_s_w = _mm256_srli_epi16(v_val_w, bits - 1);
77
1.37M
  return _mm256_avg_epu16(v_s_w, _mm256_setzero_si256());
78
1.37M
}
blend_a64_mask_avx2.c:yy_roundn_epu16
Line
Count
Source
75
1.37M
static INLINE __m256i yy_roundn_epu16(__m256i v_val_w, int bits) {
76
1.37M
  const __m256i v_s_w = _mm256_srli_epi16(v_val_w, bits - 1);
77
1.37M
  return _mm256_avg_epu16(v_s_w, _mm256_setzero_si256());
78
1.37M
}
Unexecuted instantiation: jnt_convolve_avx2.c:yy_roundn_epu16
Unexecuted instantiation: reconinter_avx2.c:yy_roundn_epu16
Unexecuted instantiation: selfguided_avx2.c:yy_roundn_epu16
Unexecuted instantiation: wiener_convolve_avx2.c:yy_roundn_epu16
Unexecuted instantiation: highbd_wiener_convolve_avx2.c:yy_roundn_epu16
79
#endif  // AOM_AOM_DSP_X86_SYNONYMS_AVX2_H_