CoCalc -- fast

CoCalc provides the best real-time collaborative environment for Jupyter Notebooks, LaTeX documents, and SageMath, scalable from individual users to large groups and classes!

GitHub Repository: hrydgard/ppsspp
Path: blob/master/Common/Math/fast/fast_matrix.c
Views: ¹⁴⁰¹
1
#include "ppsspp_config.h"
2

3
#include "Common/Math/CrossSIMD.h"
4

5
#include "fast_matrix.h"
6

7
#if PPSSPP_ARCH(X86) || PPSSPP_ARCH(AMD64)
8

9
#include <emmintrin.h>
10

11
#include "fast_matrix.h"
12

13
void fast_matrix_mul_4x4_sse(float *dest, const float *a, const float *b) {
14
	int i;
15
	__m128 a_col_1 = _mm_loadu_ps(a);
16
	__m128 a_col_2 = _mm_loadu_ps(&a[4]);
17
	__m128 a_col_3 = _mm_loadu_ps(&a[8]);
18
	__m128 a_col_4 = _mm_loadu_ps(&a[12]);
19

20
	for (i = 0; i < 16; i += 4) {
21
		__m128 r_col = _mm_mul_ps(a_col_1, _mm_set1_ps(b[i]));
22
		r_col = _mm_add_ps(r_col, _mm_mul_ps(a_col_2, _mm_set1_ps(b[i + 1])));
23
		r_col = _mm_add_ps(r_col, _mm_mul_ps(a_col_3, _mm_set1_ps(b[i + 2])));
24
		r_col = _mm_add_ps(r_col, _mm_mul_ps(a_col_4, _mm_set1_ps(b[i + 3])));
25
		_mm_storeu_ps(&dest[i], r_col);
26
	}
27
}
28

29
#elif PPSSPP_ARCH(ARM_NEON)
30

31
#if defined(_MSC_VER) && PPSSPP_ARCH(ARM64)
32
#include <arm64_neon.h>
33
#else
34
#include <arm_neon.h>
35
#endif
36

37
#if PPSSPP_ARCH(ARM)
38
static inline float32x4_t vfmaq_laneq_f32(float32x4_t _s, float32x4_t _a, float32x4_t _b, int lane) {
39
	if (lane == 0)      return vmlaq_lane_f32(_s, _a, vget_low_f32(_b), 0);
40
	else if (lane == 1) return vmlaq_lane_f32(_s, _a, vget_low_f32(_b), 1);
41
	else if (lane == 2) return vmlaq_lane_f32(_s, _a, vget_high_f32(_b), 0);
42
	else if (lane == 3) return vmlaq_lane_f32(_s, _a, vget_high_f32(_b), 1);
43
	else return vdupq_n_f32(0.f);
44
}
45
#endif
46

47
// From https://developer.arm.com/documentation/102467/0100/Matrix-multiplication-example
48
void fast_matrix_mul_4x4_neon(float *C, const float *A, const float *B) {
49
	// these are the columns A
50
	float32x4_t A0;
51
	float32x4_t A1;
52
	float32x4_t A2;
53
	float32x4_t A3;
54

55
	// these are the columns B
56
	float32x4_t B0;
57
	float32x4_t B1;
58
	float32x4_t B2;
59
	float32x4_t B3;
60

61
	// these are the columns C
62
	float32x4_t C0;
63
	float32x4_t C1;
64
	float32x4_t C2;
65
	float32x4_t C3;
66

67
	A0 = vld1q_f32(A);
68
	A1 = vld1q_f32(A + 4);
69
	A2 = vld1q_f32(A + 8);
70
	A3 = vld1q_f32(A + 12);
71

72
	// Multiply accumulate in 4x1 blocks, i.e. each column in C
73
	B0 = vld1q_f32(B);
74
	C0 = vmulq_laneq_f32(A0, B0, 0);
75
	C0 = vfmaq_laneq_f32(C0, A1, B0, 1);
76
	C0 = vfmaq_laneq_f32(C0, A2, B0, 2);
77
	C0 = vfmaq_laneq_f32(C0, A3, B0, 3);
78
	vst1q_f32(C, C0);
79

80
	B1 = vld1q_f32(B + 4);
81
	C1 = vmulq_laneq_f32(A0, B1, 0);
82
	C1 = vfmaq_laneq_f32(C1, A1, B1, 1);
83
	C1 = vfmaq_laneq_f32(C1, A2, B1, 2);
84
	C1 = vfmaq_laneq_f32(C1, A3, B1, 3);
85
	vst1q_f32(C + 4, C1);
86

87
	B2 = vld1q_f32(B + 8);
88
	C2 = vmulq_laneq_f32(A0, B2, 0);
89
	C2 = vfmaq_laneq_f32(C2, A1, B2, 1);
90
	C2 = vfmaq_laneq_f32(C2, A2, B2, 2);
91
	C2 = vfmaq_laneq_f32(C2, A3, B2, 3);
92
	vst1q_f32(C + 8, C2);
93

94
	B3 = vld1q_f32(B + 12);
95
	C3 = vmulq_laneq_f32(A0, B3, 0);
96
	C3 = vfmaq_laneq_f32(C3, A1, B3, 1);
97
	C3 = vfmaq_laneq_f32(C3, A2, B3, 2);
98
	C3 = vfmaq_laneq_f32(C3, A3, B3, 3);
99
	vst1q_f32(C + 12, C3);
100
}
101

102
#else
103

104
#define xx 0
105
#define xy 1
106
#define xz 2
107
#define xw 3
108
#define yx 4
109
#define yy 5
110
#define yz 6
111
#define yw 7
112
#define zx 8
113
#define zy 9
114
#define zz 10
115
#define zw 11
116
#define wx 12
117
#define wy 13
118
#define wz 14
119
#define ww 15
120

121
void fast_matrix_mul_4x4_c(float *dest, const float *a, const float *b) {
122
	dest[xx] = b[xx] * a[xx] + b[xy] * a[yx] + b[xz] * a[zx] + b[xw] * a[wx];
123
	dest[xy] = b[xx] * a[xy] + b[xy] * a[yy] + b[xz] * a[zy] + b[xw] * a[wy];
124
	dest[xz] = b[xx] * a[xz] + b[xy] * a[yz] + b[xz] * a[zz] + b[xw] * a[wz];
125
	dest[xw] = b[xx] * a[xw] + b[xy] * a[yw] + b[xz] * a[zw] + b[xw] * a[ww];
126

127
	dest[yx] = b[yx] * a[xx] + b[yy] * a[yx] + b[yz] * a[zx] + b[yw] * a[wx];
128
	dest[yy] = b[yx] * a[xy] + b[yy] * a[yy] + b[yz] * a[zy] + b[yw] * a[wy];
129
	dest[yz] = b[yx] * a[xz] + b[yy] * a[yz] + b[yz] * a[zz] + b[yw] * a[wz];
130
	dest[yw] = b[yx] * a[xw] + b[yy] * a[yw] + b[yz] * a[zw] + b[yw] * a[ww];
131

132
	dest[zx] = b[zx] * a[xx] + b[zy] * a[yx] + b[zz] * a[zx] + b[zw] * a[wx];
133
	dest[zy] = b[zx] * a[xy] + b[zy] * a[yy] + b[zz] * a[zy] + b[zw] * a[wy];
134
	dest[zz] = b[zx] * a[xz] + b[zy] * a[yz] + b[zz] * a[zz] + b[zw] * a[wz];
135
	dest[zw] = b[zx] * a[xw] + b[zy] * a[yw] + b[zz] * a[zw] + b[zw] * a[ww];
136

137
	dest[wx] = b[wx] * a[xx] + b[wy] * a[yx] + b[wz] * a[zx] + b[ww] * a[wx];
138
	dest[wy] = b[wx] * a[xy] + b[wy] * a[yy] + b[wz] * a[zy] + b[ww] * a[wy];
139
	dest[wz] = b[wx] * a[xz] + b[wy] * a[yz] + b[wz] * a[zz] + b[ww] * a[wz];
140
	dest[ww] = b[wx] * a[xw] + b[wy] * a[yw] + b[wz] * a[zw] + b[ww] * a[ww];
141
}
142

143
#endif
144

145
CoCalc provides the best real-time collaborative environment for Jupyter Notebooks, LaTeX documents, and SageMath, scalable from individual users to large groups and classes!

Product

Resources

Company
