1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
|
#include "avx2_gemm.h"
#include "interleave.h"
#include "multiply.h"
#include <cassert>
#include <emmintrin.h>
#include <immintrin.h>
#include <tmmintrin.h>
#include <xmmintrin.h>
#include <stdint.h>
namespace intgemm {
#ifdef __AVX2__
// PREPARE A: just quantization in the same memory order.
namespace {
// Read a vector of floats, multiply them, and cast to 32-bit integer.
inline __m256i QuantizerGrab(const float *input, const __m256 quant_mult_reg) {
return _mm256_cvtps_epi32(_mm256_mul_ps(*reinterpret_cast<const __m256*>(input), quant_mult_reg));
}
class QuantizeTile16 {
public:
typedef __m256i Integer;
explicit QuantizeTile16(float mult) : mult_(_mm256_set1_ps(mult)) {}
Integer Consecutive(const float *input) {
return Tile(input, input + 8);
}
Integer ForReshape(const float *input, int cols) {
// 8 rows in the first 128-bit register, 8 in the second register.
return Tile(input, input + 8 * cols);
}
private:
__m256i Tile(const float *input0, const float *input1) {
__m256i g0 = QuantizerGrab(input0, mult_);
__m256i g1 = QuantizerGrab(input1, mult_);
__m256i packed = _mm256_packs_epi32(g0, g1);
// Reorder the packed values because Intel does 0 1 2 3 8 9 10 11 4 5 6 7 12 13 14 15.
// Technically this could be removed if the PrepareB did the same reordering internally.
return _mm256_permute4x64_epi64(packed, 0xd8 /* 0, 2, 1, 3 */);
}
const __m256 mult_;
};
} // namespace
// Just quantize everything in order.
void AVX2_16bit::Quantize(const float *input, int16_t *output, float quant_mult, int size) {
assert(size % 16 == 0);
assert(reinterpret_cast<uintptr_t>(input) % 32 == 0);
QuantizeTile16 q(quant_mult);
const float *end = input + size;
for (; input != end; input += 16, output += 16) {
*reinterpret_cast<__m256i*>(output) = q.Consecutive(input);
}
}
namespace {
/* Read 8 floats at a time from input0, input1, input2, and input3. Quantize
* them to 8-bit by multiplying with quant_mult_reg then rounding. Concatenate
* the result into one register and return it.
*/
class QuantizeTile8 {
public:
typedef __m256i Integer;
explicit QuantizeTile8(float quant_mult) : mult_(_mm256_set1_ps(quant_mult)) {}
inline __m256i Consecutive(const float *input) {
return Tile(input, input + 8, input + 16, input + 24);
}
inline __m256i ForReshape(const float *input, int cols) {
// Put higher rows in the second half of the register. These will jumble
// around in the same way then conveniently land in the right place.
return Tile(input, input + 2 * cols, input + 16 * cols, input + 18 * cols);
}
private:
inline __m256i Tile(const float *input0, const float *input1, const float *input2, const float *input3) {
// Looking at the assembly, gcc has pulled this outside the loops calling this.
const __m256i neg127 = _mm256_set1_epi8(-127);
const __m256i shuffle_param = _mm256_set_epi32(7, 3, 6, 2, 5, 1, 4, 0);
// Grab 4 registers at a time in 32-bit format.
__m256i g0 = QuantizerGrab(input0, mult_);
__m256i g1 = QuantizerGrab(input1, mult_);
__m256i g2 = QuantizerGrab(input2, mult_);
__m256i g3 = QuantizerGrab(input3, mult_);
// Pack 32-bit to 16-bit.
__m256i packed0 = _mm256_packs_epi32(g0, g1);
__m256i packed1 = _mm256_packs_epi32(g2, g3);
// Pack 16-bit to 8-bit.
__m256i packed = _mm256_packs_epi16(packed0, packed1);
// Ban -128.
packed = _mm256_max_epi8(packed, neg127);
// Currently in 0 1 2 3 8 9 10 11 16 17 18 19 24 25 26 27 4 5 6 7 12 13 14 15 20 21 22 23 28 29 30 31
// Or as 32-bit integers 0 2 4 6 1 3 5 7
// Technically this could be removed so long as the rows are bigger than 16
// and the values are only used for GEMM.
return _mm256_permutevar8x32_epi32(packed, shuffle_param);
}
const __m256 mult_;
};
} // namespace
// Just quantize everything in order.
void AVX2_8bit::Quantize(const float *input, int8_t *output, float quant_mult, int size) {
assert(size % 32 == 0);
assert(reinterpret_cast<uintptr_t>(input) % 32 == 0);
QuantizeTile8 q(quant_mult);
const float *end = input + size;
for (; input != end; input += 32, output += 32) {
*reinterpret_cast<__m256i*>(output) = q.Consecutive(input);
}
}
void AVX2_16bit::PrepareB(const float *input, int16_t *output, float quant_mult, int rows, int cols) {
PrepareBFor16(input, output, QuantizeTile16(quant_mult), rows, cols);
}
void AVX2_8bit::PrepareB(const float *input, int8_t *output, float quant_mult, int rows, int cols) {
PrepareBFor8(input, output, QuantizeTile8(quant_mult), rows, cols);
}
void AVX2_16bit::Multiply(const int16_t *A, const int16_t *B, float *C, float unquant_mult, int A_rows, int width, int B_cols) {
Multiply16<__m256i, __m256>(A, B, C, unquant_mult, A_rows, width, B_cols);
}
void AVX2_8bit::Multiply(const int8_t *A, const int8_t *B, float *C, float unquant_mult, int A_rows, int width, int B_cols) {
Multiply8_SSE2OrAVX2<__m256i, __m256>(A, B, C, unquant_mult, A_rows, width, B_cols);
}
const char *const AVX2_16bit::kName = "16-bit AVX2";
const char *const AVX2_8bit::kName = "8-bit AVX2";
#endif // __AVX2__
} // namespace intgemm
|
