Coverage Report

Created: 2026-08-13 06:29

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
/src/zlib-ng/arch/x86/adler32_avx2.c
Line
Count
Source
1
/* adler32_avx2.c -- compute the Adler-32 checksum of a data stream
2
 * Copyright (C) 1995-2011 Mark Adler
3
 * Copyright (C) 2022 Adam Stylinski
4
 * Authors:
5
 *   Brian Bockelman <bockelman@gmail.com>
6
 *   Adam Stylinski <kungfujesus06@gmail.com>
7
 * For conditions of distribution and use, see copyright notice in zlib.h
8
 */
9
10
#ifdef X86_AVX2
11
12
#include "zbuild.h"
13
#include <immintrin.h>
14
#include "adler32_p.h"
15
#include "adler32_avx2_p.h"
16
#include "x86_intrins.h"
17
18
extern uint32_t adler32_copy_sse42(uint32_t adler, uint8_t *dst, const uint8_t *src, size_t len);
19
extern uint32_t adler32_ssse3(uint32_t adler, const uint8_t *src, size_t len);
20
21
1.56M
Z_FORCEINLINE static uint32_t adler32_copy_impl(uint32_t adler, uint8_t *dst, const uint8_t *src, size_t len, const int COPY) {
22
1.56M
    uint32_t adler0, adler1;
23
1.56M
    adler1 = (adler >> 16) & 0xffff;
24
1.56M
    adler0 = adler & 0xffff;
25
26
1.84M
rem_peel:
27
1.84M
    if (len < 16) {
28
1.14M
        return adler32_copy_tail(adler0, dst, src, len, adler1, 1, 15, COPY);
29
1.14M
    } else if (len < 32) {
30
375k
        if (COPY) {
31
0
            return adler32_copy_sse42(adler, dst, src, len);
32
375k
        } else {
33
375k
            return adler32_ssse3(adler, src, len);
34
375k
        }
35
375k
    }
36
37
325k
    __m256i vs1, vs2, vs2_0;
38
39
325k
    const __m256i dot2v = _mm256_setr_epi8(64, 63, 62, 61, 60, 59, 58, 57, 56, 55, 54, 53, 52, 51, 50, 49, 48, 47,
40
325k
                                           46, 45, 44, 43, 42, 41, 40, 39, 38, 37, 36, 35, 34, 33);
41
325k
    const __m256i dot2v_0 = _mm256_setr_epi8(32, 31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, 18, 17, 16, 15,
42
325k
                                             14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1);
43
325k
    const __m256i dot3v = _mm256_set1_epi16(1);
44
325k
    const __m256i zero = _mm256_setzero_si256();
45
46
659k
    while (len >= 32) {
47
333k
        vs1 = _mm256_zextsi128_si256(_mm_cvtsi32_si128(adler0));
48
333k
        vs2 = _mm256_zextsi128_si256(_mm_cvtsi32_si128(adler1));
49
333k
        __m256i vs1_0 = vs1;
50
333k
        __m256i vs3 = _mm256_setzero_si256();
51
333k
        vs2_0 = vs3;
52
53
333k
        size_t k = ALIGN_DOWN(MIN(len, NMAX), 32);
54
333k
        len -= k;
55
56
1.40M
        while (k >= 64) {
57
1.06M
            __m256i vbuf = _mm256_loadu_si256((__m256i*)src);
58
1.06M
            __m256i vbuf_0 = _mm256_loadu_si256((__m256i*)(src + 32));
59
1.06M
            src += 64;
60
1.06M
            k -= 64;
61
62
1.06M
            __m256i vs1_sad = _mm256_sad_epu8(vbuf, zero);
63
1.06M
            __m256i vs1_sad2 = _mm256_sad_epu8(vbuf_0, zero);
64
65
1.06M
            if (COPY) {
66
0
                _mm256_storeu_si256((__m256i*)dst, vbuf);
67
0
                _mm256_storeu_si256((__m256i*)(dst + 32), vbuf_0);
68
0
                dst += 64;
69
0
            }
70
71
1.06M
            vs1 = _mm256_add_epi32(vs1, vs1_sad);
72
1.06M
            vs3 = _mm256_add_epi32(vs3, vs1_0);
73
1.06M
            __m256i v_short_sum2 = _mm256_maddubs_epi16(vbuf, dot2v); // sum 32 uint8s to 16 shorts
74
1.06M
            __m256i v_short_sum2_0 = _mm256_maddubs_epi16(vbuf_0, dot2v_0); // sum 32 uint8s to 16 shorts
75
1.06M
            __m256i vsum2 = _mm256_madd_epi16(v_short_sum2, dot3v); // sum 16 shorts to 8 uint32s
76
1.06M
            __m256i vsum2_0 = _mm256_madd_epi16(v_short_sum2_0, dot3v); // sum 16 shorts to 8 uint32s
77
1.06M
            vs1 = _mm256_add_epi32(vs1_sad2, vs1);
78
1.06M
            vs2 = _mm256_add_epi32(vsum2, vs2);
79
1.06M
            vs2_0 = _mm256_add_epi32(vsum2_0, vs2_0);
80
1.06M
            vs1_0 = vs1;
81
1.06M
        }
82
83
333k
        vs2 = _mm256_add_epi32(vs2_0, vs2);
84
333k
        vs3 = _mm256_slli_epi32(vs3, 6);
85
333k
        vs2 = _mm256_add_epi32(vs3, vs2);
86
333k
        vs3 = _mm256_setzero_si256();
87
88
625k
        while (k >= 32) {
89
            /*
90
               vs1 = adler + sum(c[i])
91
               vs2 = sum2 + 32 vs1 + sum( (32-i+1) c[i] )
92
            */
93
291k
            __m256i vbuf = _mm256_loadu_si256((__m256i*)src);
94
291k
            src += 32;
95
291k
            k -= 32;
96
97
291k
            __m256i vs1_sad = _mm256_sad_epu8(vbuf, zero); // Sum of abs diff, resulting in 2 x int32's
98
99
291k
            if (COPY) {
100
0
                _mm256_storeu_si256((__m256i*)dst, vbuf);
101
0
                dst += 32;
102
0
            }
103
104
291k
            vs1 = _mm256_add_epi32(vs1, vs1_sad);
105
291k
            vs3 = _mm256_add_epi32(vs3, vs1_0);
106
291k
            __m256i v_short_sum2 = _mm256_maddubs_epi16(vbuf, dot2v_0); // sum 32 uint8s to 16 shorts
107
291k
            __m256i vsum2 = _mm256_madd_epi16(v_short_sum2, dot3v); // sum 16 shorts to 8 uint32s
108
291k
            vs2 = _mm256_add_epi32(vsum2, vs2);
109
291k
            vs1_0 = vs1;
110
291k
        }
111
112
        /* Defer the multiplication with 32 to outside of the loop */
113
333k
        vs3 = _mm256_slli_epi32(vs3, 5);
114
333k
        vs2 = _mm256_add_epi32(vs2, vs3);
115
116
        /* The compiler is generating the following sequence for this integer modulus
117
         * when done the scalar way, in GPRs:
118
119
         adler = (s1_unpack[0] % BASE) + (s1_unpack[1] % BASE) + (s1_unpack[2] % BASE) + (s1_unpack[3] % BASE) +
120
                 (s1_unpack[4] % BASE) + (s1_unpack[5] % BASE) + (s1_unpack[6] % BASE) + (s1_unpack[7] % BASE);
121
122
         mov    $0x80078071,%edi // move magic constant into 32 bit register %edi
123
         ...
124
         vmovd  %xmm1,%esi // move vector lane 0 to 32 bit register %esi
125
         mov    %rsi,%rax  // zero-extend this value to 64 bit precision in %rax
126
         imul   %rdi,%rsi // do a signed multiplication with magic constant and vector element
127
         shr    $0x2f,%rsi // shift right by 47
128
         imul   $0xfff1,%esi,%esi // do a signed multiplication with value truncated to 32 bits with 0xfff1
129
         sub    %esi,%eax // subtract lower 32 bits of original vector value from modified one above
130
         ...
131
         // repeats for each element with vpextract instructions
132
133
         This is tricky with AVX2 for a number of reasons:
134
             1.) There's no 64 bit multiplication instruction, but there is a sequence to get there
135
             2.) There's ways to extend vectors to 64 bit precision, but no simple way to truncate
136
                 back down to 32 bit precision later (there is in AVX512)
137
             3.) Full width integer multiplications aren't cheap
138
139
         We can, however, do a relatively cheap sequence for horizontal sums.
140
         Then, we simply do the integer modulus on the resulting 64 bit GPR, on a scalar value. It was
141
         previously thought that casting to 64 bit precision was needed prior to the horizontal sum, but
142
         that is simply not the case, as NMAX is defined as the maximum number of scalar sums that can be
143
         performed on the maximum possible inputs before overflow
144
         */
145
146
147
         /* In AVX2-land, this trip through GPRs will probably be unavoidable, as there's no cheap and easy
148
          * conversion from 64 bit integer to 32 bit (needed for the inexpensive modulus with a constant).
149
          * This casting to 32 bit is cheap through GPRs (just register aliasing). See above for exactly
150
          * what the compiler is doing to avoid integer divisions. */
151
333k
         adler0 = partial_hsum256(vs1) % BASE;
152
333k
         adler1 = hsum256(vs2) % BASE;
153
333k
    }
154
155
325k
    adler = adler0 | (adler1 << 16);
156
157
325k
    if (len) {
158
283k
        goto rem_peel;
159
283k
    }
160
161
42.6k
    return adler;
162
325k
}
163
164
1.56M
Z_INTERNAL uint32_t adler32_avx2(uint32_t adler, const uint8_t *src, size_t len) {
165
1.56M
    return adler32_copy_impl(adler, NULL, src, len, 0);
166
1.56M
}
167
168
0
Z_INTERNAL uint32_t adler32_copy_avx2(uint32_t adler, uint8_t *dst, const uint8_t *src, size_t len) {
169
0
    return adler32_copy_impl(adler, dst, src, len, 1);
170
0
}
171
172
#endif