1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_dot_prod_f16.c
4 * Description: Floating-point dot product
5 *
6 * $Date: 23 April 2021
7 * $Revision: V1.9.0
8 *
9 * Target Processor: Cortex-M and Cortex-A cores
10 * -------------------------------------------------------------------- */
11 /*
12 * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved.
13 *
14 * SPDX-License-Identifier: Apache-2.0
15 *
16 * Licensed under the Apache License, Version 2.0 (the License); you may
17 * not use this file except in compliance with the License.
18 * You may obtain a copy of the License at
19 *
20 * www.apache.org/licenses/LICENSE-2.0
21 *
22 * Unless required by applicable law or agreed to in writing, software
23 * distributed under the License is distributed on an AS IS BASIS, WITHOUT
24 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
25 * See the License for the specific language governing permissions and
26 * limitations under the License.
27 */
28
29 #include "dsp/basic_math_functions_f16.h"
30
31 /**
32 @ingroup groupMath
33 */
34
35 /**
36 @defgroup BasicDotProd Vector Dot Product
37
38 Computes the dot product of two vectors.
39 The vectors are multiplied element-by-element and then summed.
40
41 <pre>
42 sum = pSrcA[0]*pSrcB[0] + pSrcA[1]*pSrcB[1] + ... + pSrcA[blockSize-1]*pSrcB[blockSize-1]
43 </pre>
44
45 There are separate functions for floating-point, Q7, Q15, and Q31 data types.
46 */
47
48 /**
49 @addtogroup BasicDotProd
50 @{
51 */
52
53 /**
54 @brief Dot product of floating-point vectors.
55 @param[in] pSrcA points to the first input vector.
56 @param[in] pSrcB points to the second input vector.
57 @param[in] blockSize number of samples in each vector.
58 @param[out] result output result returned here.
59 @return none
60 */
61
62 #if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
63
64 #include "arm_helium_utils.h"
65
66
arm_dot_prod_f16(const float16_t * pSrcA,const float16_t * pSrcB,uint32_t blockSize,float16_t * result)67 void arm_dot_prod_f16(
68 const float16_t * pSrcA,
69 const float16_t * pSrcB,
70 uint32_t blockSize,
71 float16_t * result)
72 {
73 f16x8_t vecA, vecB;
74 f16x8_t vecSum;
75 uint32_t blkCnt;
76 float16_t sum = 0.0f;
77 vecSum = vdupq_n_f16(0.0f);
78
79 /* Compute 4 outputs at a time */
80 blkCnt = blockSize >> 3U;
81 while (blkCnt > 0U)
82 {
83 /*
84 * C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1]
85 * Calculate dot product and then store the result in a temporary buffer.
86 * and advance vector source and destination pointers
87 */
88 vecA = vld1q(pSrcA);
89 pSrcA += 8;
90
91 vecB = vld1q(pSrcB);
92 pSrcB += 8;
93
94 vecSum = vfmaq(vecSum, vecA, vecB);
95 /*
96 * Decrement the blockSize loop counter
97 */
98 blkCnt --;
99 }
100
101
102 blkCnt = blockSize & 7;
103 if (blkCnt > 0U)
104 {
105 /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */
106
107 mve_pred16_t p0 = vctp16q(blkCnt);
108 vecA = vld1q(pSrcA);
109 vecB = vld1q(pSrcB);
110 vecSum = vfmaq_m(vecSum, vecA, vecB, p0);
111 }
112
113 sum = vecAddAcrossF16Mve(vecSum);
114
115 /* Store result in destination buffer */
116 *result = sum;
117
118 }
119
120 #else
121 #if defined(ARM_FLOAT16_SUPPORTED)
arm_dot_prod_f16(const float16_t * pSrcA,const float16_t * pSrcB,uint32_t blockSize,float16_t * result)122 void arm_dot_prod_f16(
123 const float16_t * pSrcA,
124 const float16_t * pSrcB,
125 uint32_t blockSize,
126 float16_t * result)
127 {
128 uint32_t blkCnt; /* Loop counter */
129 _Float16 sum = 0.0f; /* Temporary return variable */
130
131
132 #if defined (ARM_MATH_LOOPUNROLL) && !defined(ARM_MATH_AUTOVECTORIZE)
133
134 /* Loop unrolling: Compute 4 outputs at a time */
135 blkCnt = blockSize >> 2U;
136
137 /* First part of the processing with loop unrolling. Compute 4 outputs at a time.
138 ** a second loop below computes the remaining 1 to 3 samples. */
139 while (blkCnt > 0U)
140 {
141 /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */
142
143 /* Calculate dot product and store result in a temporary buffer. */
144 sum += (_Float16)(*pSrcA++) * (_Float16)(*pSrcB++);
145
146 sum += (_Float16)(*pSrcA++) * (_Float16)(*pSrcB++);
147
148 sum += (_Float16)(*pSrcA++) * (_Float16)(*pSrcB++);
149
150 sum += (_Float16)(*pSrcA++) * (_Float16)(*pSrcB++);
151
152 /* Decrement loop counter */
153 blkCnt--;
154 }
155
156 /* Loop unrolling: Compute remaining outputs */
157 blkCnt = blockSize % 0x4U;
158
159 #else
160
161 /* Initialize blkCnt with number of samples */
162 blkCnt = blockSize;
163
164 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
165
166 while (blkCnt > 0U)
167 {
168 /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */
169
170 /* Calculate dot product and store result in a temporary buffer. */
171 sum += (_Float16)(*pSrcA++) * (_Float16)(*pSrcB++);
172
173 /* Decrement loop counter */
174 blkCnt--;
175 }
176
177 /* Store result in destination buffer */
178 *result = sum;
179 }
180 #endif
181 #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
182 /**
183 @} end of BasicDotProd group
184 */
185
