1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_mat_add_f16.c
4 * Description: Floating-point matrix addition
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/matrix_functions_f16.h"
30
31 #if defined(ARM_FLOAT16_SUPPORTED)
32
33
34 /**
35 @ingroup groupMatrix
36 */
37
38
39 /**
40 @addtogroup MatrixAdd
41 @{
42 */
43
44
45 /**
46 @brief Floating-point matrix addition.
47 @param[in] pSrcA points to first input matrix structure
48 @param[in] pSrcB points to second input matrix structure
49 @param[out] pDst points to output matrix structure
50 @return execution status
51 - \ref ARM_MATH_SUCCESS : Operation successful
52 - \ref ARM_MATH_SIZE_MISMATCH : Matrix size check failed
53 */
54
55 #if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
56
arm_mat_add_f16(const arm_matrix_instance_f16 * pSrcA,const arm_matrix_instance_f16 * pSrcB,arm_matrix_instance_f16 * pDst)57 arm_status arm_mat_add_f16(
58 const arm_matrix_instance_f16 * pSrcA,
59 const arm_matrix_instance_f16 * pSrcB,
60 arm_matrix_instance_f16 * pDst)
61 {
62 arm_status status;
63 uint32_t numSamples; /* total number of elements in the matrix */
64 float16_t *pDataA, *pDataB, *pDataDst;
65 f16x8_t vecA, vecB, vecDst;
66 float16_t const *pSrcAVec;
67 float16_t const *pSrcBVec;
68 uint32_t blkCnt; /* loop counters */
69
70 pDataA = pSrcA->pData;
71 pDataB = pSrcB->pData;
72 pDataDst = pDst->pData;
73 pSrcAVec = (float16_t const *) pDataA;
74 pSrcBVec = (float16_t const *) pDataB;
75
76 #ifdef ARM_MATH_MATRIX_CHECK
77 /* Check for matrix mismatch condition */
78 if ((pSrcA->numRows != pSrcB->numRows) ||
79 (pSrcA->numCols != pSrcB->numCols) ||
80 (pSrcA->numRows != pDst->numRows) || (pSrcA->numCols != pDst->numCols))
81 {
82 /* Set status as ARM_MATH_SIZE_MISMATCH */
83 status = ARM_MATH_SIZE_MISMATCH;
84 }
85 else
86 #endif
87 {
88 /*
89 * Total number of samples in the input matrix
90 */
91 numSamples = (uint32_t) pSrcA->numRows * pSrcA->numCols;
92 blkCnt = numSamples >> 3;
93 while (blkCnt > 0U)
94 {
95 /* C(m,n) = A(m,n) + B(m,n) */
96 /* Add and then store the results in the destination buffer. */
97 vecA = vld1q(pSrcAVec);
98 pSrcAVec += 8;
99 vecB = vld1q(pSrcBVec);
100 pSrcBVec += 8;
101 vecDst = vaddq(vecA, vecB);
102 vst1q(pDataDst, vecDst);
103 pDataDst += 8;
104 /*
105 * Decrement the blockSize loop counter
106 */
107 blkCnt--;
108 }
109 /*
110 * tail
111 */
112 blkCnt = numSamples & 7;
113 if (blkCnt > 0U)
114 {
115 mve_pred16_t p0 = vctp16q(blkCnt);
116 vecA = vld1q(pSrcAVec);
117 vecB = vld1q(pSrcBVec);
118 vecDst = vaddq_m(vecDst, vecA, vecB, p0);
119 vstrhq_p(pDataDst, vecDst, p0);
120 }
121 /* set status as ARM_MATH_SUCCESS */
122 status = ARM_MATH_SUCCESS;
123 }
124 return (status);
125 }
126 #else
127
arm_mat_add_f16(const arm_matrix_instance_f16 * pSrcA,const arm_matrix_instance_f16 * pSrcB,arm_matrix_instance_f16 * pDst)128 arm_status arm_mat_add_f16(
129 const arm_matrix_instance_f16 * pSrcA,
130 const arm_matrix_instance_f16 * pSrcB,
131 arm_matrix_instance_f16 * pDst)
132 {
133 float16_t *pInA = pSrcA->pData; /* input data matrix pointer A */
134 float16_t *pInB = pSrcB->pData; /* input data matrix pointer B */
135 float16_t *pOut = pDst->pData; /* output data matrix pointer */
136
137 uint32_t numSamples; /* total number of elements in the matrix */
138 uint32_t blkCnt; /* loop counters */
139 arm_status status; /* status of matrix addition */
140
141 #ifdef ARM_MATH_MATRIX_CHECK
142
143 /* Check for matrix mismatch condition */
144 if ((pSrcA->numRows != pSrcB->numRows) ||
145 (pSrcA->numCols != pSrcB->numCols) ||
146 (pSrcA->numRows != pDst->numRows) ||
147 (pSrcA->numCols != pDst->numCols) )
148 {
149 /* Set status as ARM_MATH_SIZE_MISMATCH */
150 status = ARM_MATH_SIZE_MISMATCH;
151 }
152 else
153
154 #endif /* #ifdef ARM_MATH_MATRIX_CHECK */
155
156 {
157 /* Total number of samples in input matrix */
158 numSamples = (uint32_t) pSrcA->numRows * pSrcA->numCols;
159
160 #if defined (ARM_MATH_LOOPUNROLL)
161
162 /* Loop unrolling: Compute 4 outputs at a time */
163 blkCnt = numSamples >> 2U;
164
165 while (blkCnt > 0U)
166 {
167 /* C(m,n) = A(m,n) + B(m,n) */
168
169 /* Add and store result in destination buffer. */
170 *pOut++ = *pInA++ + *pInB++;
171
172 *pOut++ = *pInA++ + *pInB++;
173
174 *pOut++ = *pInA++ + *pInB++;
175
176 *pOut++ = *pInA++ + *pInB++;
177
178 /* Decrement loop counter */
179 blkCnt--;
180 }
181
182 /* Loop unrolling: Compute remaining outputs */
183 blkCnt = numSamples % 0x4U;
184
185 #else
186
187 /* Initialize blkCnt with number of samples */
188 blkCnt = numSamples;
189
190 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
191
192 while (blkCnt > 0U)
193 {
194 /* C(m,n) = A(m,n) + B(m,n) */
195
196 /* Add and store result in destination buffer. */
197 *pOut++ = *pInA++ + *pInB++;
198
199 /* Decrement loop counter */
200 blkCnt--;
201 }
202
203 /* Set status as ARM_MATH_SUCCESS */
204 status = ARM_MATH_SUCCESS;
205 }
206
207 /* Return to application */
208 return (status);
209 }
210 #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
211
212 /**
213 @} end of MatrixAdd group
214 */
215
216 #endif /* #if defined(ARM_FLOAT16_SUPPORTED) */
217
218