1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_sub_f16.c
4 * Description: Floating-point vector subtraction
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 /**
37 @addtogroup BasicSub
38 @{
39 */
40
41 /**
42 @brief Floating-point vector subtraction.
43 @param[in] pSrcA points to the first input vector
44 @param[in] pSrcB points to the second input vector
45 @param[out] pDst points to the output vector
46 @param[in] blockSize number of samples in each vector
47 */
48
49 #if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
50 #include "arm_helium_utils.h"
51
arm_sub_f16(const float16_t * pSrcA,const float16_t * pSrcB,float16_t * pDst,uint32_t blockSize)52 void arm_sub_f16(
53 const float16_t * pSrcA,
54 const float16_t * pSrcB,
55 float16_t * pDst,
56 uint32_t blockSize)
57 {
58 uint32_t blkCnt; /* Loop counter */
59
60 f16x8_t vec1;
61 f16x8_t vec2;
62 f16x8_t res;
63
64 /* Compute 4 outputs at a time */
65 blkCnt = blockSize >> 3U;
66
67 while (blkCnt > 0U)
68 {
69 /* C = A + B */
70
71 /* Add and then store the results in the destination buffer. */
72 vec1 = vld1q(pSrcA);
73 vec2 = vld1q(pSrcB);
74 res = vsubq(vec1, vec2);
75 vst1q(pDst, res);
76
77 /* Increment pointers */
78 pSrcA += 8;
79 pSrcB += 8;
80 pDst += 8;
81
82 /* Decrement the loop counter */
83 blkCnt--;
84 }
85
86 /* Tail */
87 blkCnt = blockSize & 0x7;
88
89 if (blkCnt > 0U)
90 {
91 /* C = A + B */
92 mve_pred16_t p0 = vctp16q(blkCnt);
93 vec1 = vld1q(pSrcA);
94 vec2 = vld1q(pSrcB);
95 vstrhq_p(pDst, vsubq(vec1,vec2), p0);
96 }
97
98 }
99
100 #else
101 #if defined(ARM_FLOAT16_SUPPORTED)
arm_sub_f16(const float16_t * pSrcA,const float16_t * pSrcB,float16_t * pDst,uint32_t blockSize)102 void arm_sub_f16(
103 const float16_t * pSrcA,
104 const float16_t * pSrcB,
105 float16_t * pDst,
106 uint32_t blockSize)
107 {
108 uint32_t blkCnt; /* Loop counter */
109
110 #if defined (ARM_MATH_LOOPUNROLL) && !defined(ARM_MATH_AUTOVECTORIZE)
111
112 /* Loop unrolling: Compute 4 outputs at a time */
113 blkCnt = blockSize >> 2U;
114
115 while (blkCnt > 0U)
116 {
117 /* C = A - B */
118
119 /* Subtract and store result in destination buffer. */
120 *pDst++ = (_Float16)(*pSrcA++) - (_Float16)(*pSrcB++);
121
122 *pDst++ = (_Float16)(*pSrcA++) - (_Float16)(*pSrcB++);
123
124 *pDst++ = (_Float16)(*pSrcA++) - (_Float16)(*pSrcB++);
125
126 *pDst++ = (_Float16)(*pSrcA++) - (_Float16)(*pSrcB++);
127
128 /* Decrement loop counter */
129 blkCnt--;
130 }
131
132 /* Loop unrolling: Compute remaining outputs */
133 blkCnt = blockSize % 0x4U;
134
135 #else
136
137 /* Initialize blkCnt with number of samples */
138 blkCnt = blockSize;
139
140 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
141
142 while (blkCnt > 0U)
143 {
144 /* C = A - B */
145
146 /* Subtract and store result in destination buffer. */
147 *pDst++ = (_Float16)(*pSrcA++) - (_Float16)(*pSrcB++);
148
149 /* Decrement loop counter */
150 blkCnt--;
151 }
152
153 }
154 #endif
155 #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
156
157 /**
158 @} end of BasicSub group
159 */
160
