1 /* ----------------------------------------------------------------------
2  * Project:      CMSIS DSP Library
3  * Title:        arm_float_to_q15.c
4  * Description:  Converts the elements of the floating-point vector to Q15 vector
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/support_functions_f16.h"
30 
31 #if defined(ARM_FLOAT16_SUPPORTED)
32 
33 
34 /**
35   @ingroup groupSupport
36  */
37 
38 /**
39   @addtogroup f16_to_x
40   @{
41  */
42 
43 /**
44   @brief         Converts the elements of the f16 vector to Q15 vector.
45   @param[in]     pSrc       points to the f16 input vector
46   @param[out]    pDst       points to the Q15 output vector
47   @param[in]     blockSize  number of samples in each vector
48   @return        none
49 
50   @par           Details
51                    The equation used for the conversion process is:
52   <pre>
53       pDst[n] = (q15_t)(pSrc[n] * 32768);   0 <= n < blockSize.
54   </pre>
55 
56   @par           Scaling and Overflow Behavior
57                    The function uses saturating arithmetic.
58                    Results outside of the allowable Q15 range [0x8000 0x7FFF] are saturated.
59 
60   @note
61                    In order to apply rounding in scalar version, the library should be rebuilt with the ROUNDING macro
62                    defined in the preprocessor section of project options.
63  */
64 
65 #if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
66 
arm_f16_to_q15(const float16_t * pSrc,q15_t * pDst,uint32_t blockSize)67 void arm_f16_to_q15(
68   const float16_t * pSrc,
69   q15_t * pDst,
70   uint32_t blockSize)
71 {
72     float16_t       maxQ = (float16_t) Q15_MAX;
73     float16x8_t         vecDst;
74 
75 
76     do {
77         mve_pred16_t    p = vctp16q(blockSize);
78 
79         vecDst = vldrhq_z_f16((float16_t const *) pSrc, p);
80         /* C = A * 32767 */
81         /* convert from float to Q15 and then store the results in the destination buffer */
82         vecDst = vmulq_m(vuninitializedq_f16(), vecDst, maxQ, p);
83 
84         vstrhq_p_s16(pDst,
85             vcvtaq_m(vuninitializedq_s16(), vecDst, p), p);
86         /*
87          * Decrement the blockSize loop counter
88          * Advance vector source and destination pointers
89          */
90         pSrc += 8;
91         pDst += 8;
92         blockSize -= 8;
93     }
94     while ((int32_t) blockSize > 0);
95 }
96 
97 #else
98 
arm_f16_to_q15(const float16_t * pSrc,q15_t * pDst,uint32_t blockSize)99 void arm_f16_to_q15(
100   const float16_t * pSrc,
101         q15_t * pDst,
102         uint32_t blockSize)
103 {
104     const float16_t *pIn = pSrc;      /* Src pointer */
105     uint32_t  blkCnt;           /* loop counter */
106 #ifdef ARM_MATH_ROUNDING
107     float16_t in;
108 #endif                          /*      #ifdef ARM_MATH_ROUNDING        */
109 
110     /*
111      * Loop over blockSize number of values
112      */
113     blkCnt = blockSize;
114 
115     while (blkCnt > 0U)
116     {
117 
118 #ifdef ARM_MATH_ROUNDING
119 
120         /*
121          * C = A * 65536
122          */
123         /*
124          * convert from float to Q31 and then store the results in the destination buffer
125          */
126         in = *pIn++;
127         in = (in * 32768.0);
128         in += in > 0.0 ? 0.5 : -0.5;
129         *pDst++ = clip_q31_to_q15((q31_t) (in));
130 
131 #else
132 
133         /*
134          * C = A * 32768
135          */
136         /*
137          * convert from float to Q31 and then store the results in the destination buffer
138          */
139         *pDst++ = clip_q31_to_q15((q31_t) (*pIn++ * 32768.0));
140 
141 #endif                          /*      #ifdef ARM_MATH_ROUNDING        */
142 
143         /*
144          * Decrement the loop counter
145          */
146         blkCnt--;
147     }
148 
149 }
150 #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
151 
152 /**
153   @} end of f16_to_x group
154  */
155 
156 #endif /* #if defined(ARM_FLOAT16_SUPPORTED) */
157 
158