GCC Code Coverage Report

Directory: ./
File: kernels/volk/volk_32f_64f_multiply_64f.h
Date: 2023-10-23 23:10:04
Exec Total Coverage
Lines: 56 56 100.0%
Functions: 3 3 100.0%
Branches: 10 10 100.0%
Line Branch Exec Source
1 /* -*- c++ -*- */
2 /*
3 * Copyright 2018 Free Software Foundation, Inc.
4 *
5 * This file is part of VOLK
6 *
7 * SPDX-License-Identifier: LGPL-3.0-or-later
8 */
9
10 /*!
11 * \page volk_32f_64f_multiply_64f
12 *
13 * \b Overview
14 *
15 * Multiplies two input double-precision doubleing point vectors together.
16 *
17 * c[i] = a[i] * b[i]
18 *
19 * <b>Dispatcher Prototype</b>
20 * \code
21 * void volk_32f_64f_multiply_64f(double* cVector, const double* aVector, const double*
22 * bVector, unsigned int num_points) \endcode
23 *
24 * \b Inputs
25 * \li aVector: First input vector.
26 * \li bVector: Second input vector.
27 * \li num_points: The number of values in both input vectors.
28 *
29 * \b Outputs
30 * \li cVector: The output vector.
31 *
32 * \b Example
33 * Multiply elements of an increasing vector by those of a decreasing vector.
34 * \code
35 * int N = 10;
36 * unsigned int alignment = volk_get_alignment();
37 * float* increasing = (float*)volk_malloc(sizeof(float)*N, alignment);
38 * double* decreasing = (double*)volk_malloc(sizeof(double)*N, alignment);
39 * double* out = (double*)volk_malloc(sizeof(double)*N, alignment);
40 *
41 * for(unsigned int ii = 0; ii < N; ++ii){
42 * increasing[ii] = (double)ii;
43 * decreasing[ii] = 10.f - (double)ii;
44 * }
45 *
46 * volk_32f_64f_multiply_64f(out, increasing, decreasing, N);
47 *
48 * for(unsigned int ii = 0; ii < N; ++ii){
49 * printf("out[%u] = %1.2F\n", ii, out[ii]);
50 * }
51 *
52 * volk_free(increasing);
53 * volk_free(decreasing);
54 * volk_free(out);
55 * \endcode
56 */
57
58 #ifndef INCLUDED_volk_32f_64f_multiply_64f_H
59 #define INCLUDED_volk_32f_64f_multiply_64f_H
60
61 #include <inttypes.h>
62
63
64 #ifdef LV_HAVE_GENERIC
65
66 2 static inline void volk_32f_64f_multiply_64f_generic(double* cVector,
67 const float* aVector,
68 const double* bVector,
69 unsigned int num_points)
70 {
71 2 double* cPtr = cVector;
72 2 const float* aPtr = aVector;
73 2 const double* bPtr = bVector;
74 2 unsigned int number = 0;
75
76
2/2
✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
 262144 for (number = 0; number < num_points; number++) {
77 262142 *cPtr++ = ((double)(*aPtr++)) * (*bPtr++);
78 }
79 2 }
80
81 #endif /* LV_HAVE_GENERIC */
82
83 /*
84 * Unaligned versions
85 */
86
87
88 #ifdef LV_HAVE_AVX
89
90 #include <immintrin.h>
91 #include <xmmintrin.h>
92
93 2 static inline void volk_32f_64f_multiply_64f_u_avx(double* cVector,
94 const float* aVector,
95 const double* bVector,
96 unsigned int num_points)
97 {
98 2 unsigned int number = 0;
99 2 const unsigned int eighth_points = num_points / 8;
100
101 2 double* cPtr = cVector;
102 2 const float* aPtr = aVector;
103 2 const double* bPtr = bVector;
104
105 __m256 aVal;
106 __m128 aVal1, aVal2;
107 __m256d aDbl1, aDbl2, bVal1, bVal2, cVal1, cVal2;
108
2/2
✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
 32768 for (; number < eighth_points; number++) {
109
110 32766 aVal = _mm256_loadu_ps(aPtr);
111 32766 bVal1 = _mm256_loadu_pd(bPtr);
112 32766 bVal2 = _mm256_loadu_pd(bPtr + 4);
113
114 32766 aVal1 = _mm256_extractf128_ps(aVal, 0);
115 32766 aVal2 = _mm256_extractf128_ps(aVal, 1);
116
117 32766 aDbl1 = _mm256_cvtps_pd(aVal1);
118 32766 aDbl2 = _mm256_cvtps_pd(aVal2);
119
120 32766 cVal1 = _mm256_mul_pd(aDbl1, bVal1);
121 32766 cVal2 = _mm256_mul_pd(aDbl2, bVal2);
122
123 _mm256_storeu_pd(cPtr, cVal1); // Store the results back into the C container
124 32766 _mm256_storeu_pd(cPtr + 4, cVal2); // Store the results back into the C container
125
126 32766 aPtr += 8;
127 32766 bPtr += 8;
128 32766 cPtr += 8;
129 }
130
131 2 number = eighth_points * 8;
132
2/2
✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
 16 for (; number < num_points; number++) {
133 14 *cPtr++ = ((double)(*aPtr++)) * (*bPtr++);
134 }
135 2 }
136
137 #endif /* LV_HAVE_AVX */
138
139
140 #ifdef LV_HAVE_AVX
141
142 #include <immintrin.h>
143 #include <xmmintrin.h>
144
145 2 static inline void volk_32f_64f_multiply_64f_a_avx(double* cVector,
146 const float* aVector,
147 const double* bVector,
148 unsigned int num_points)
149 {
150 2 unsigned int number = 0;
151 2 const unsigned int eighth_points = num_points / 8;
152
153 2 double* cPtr = cVector;
154 2 const float* aPtr = aVector;
155 2 const double* bPtr = bVector;
156
157 __m256 aVal;
158 __m128 aVal1, aVal2;
159 __m256d aDbl1, aDbl2, bVal1, bVal2, cVal1, cVal2;
160
2/2
✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
 32768 for (; number < eighth_points; number++) {
161
162 32766 aVal = _mm256_load_ps(aPtr);
163 32766 bVal1 = _mm256_load_pd(bPtr);
164 32766 bVal2 = _mm256_load_pd(bPtr + 4);
165
166 32766 aVal1 = _mm256_extractf128_ps(aVal, 0);
167 32766 aVal2 = _mm256_extractf128_ps(aVal, 1);
168
169 32766 aDbl1 = _mm256_cvtps_pd(aVal1);
170 32766 aDbl2 = _mm256_cvtps_pd(aVal2);
171
172 32766 cVal1 = _mm256_mul_pd(aDbl1, bVal1);
173 32766 cVal2 = _mm256_mul_pd(aDbl2, bVal2);
174
175 _mm256_store_pd(cPtr, cVal1); // Store the results back into the C container
176 32766 _mm256_store_pd(cPtr + 4, cVal2); // Store the results back into the C container
177
178 32766 aPtr += 8;
179 32766 bPtr += 8;
180 32766 cPtr += 8;
181 }
182
183 2 number = eighth_points * 8;
184
2/2
✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
 16 for (; number < num_points; number++) {
185 14 *cPtr++ = ((double)(*aPtr++)) * (*bPtr++);
186 }
187 2 }
188
189 #endif /* LV_HAVE_AVX */
190
191
192 #endif /* INCLUDED_volk_32f_64f_multiply_64f_u_H */
193