1 /* ----------------------------------------------------------------------
2  * Project:      CMSIS DSP Library
3  * Title:        arm_q15_to_float.c
4  * Description:  Converts the elements of the Q15 vector to floating-point 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  * @defgroup q15_to_x  Convert 16-bit fixed point value
40  */
41 
42 /**
43   @addtogroup q15_to_x
44   @{
45  */
46 
47 /**
48   @brief         Converts the elements of the Q15 vector to f16 vector.
49   @param[in]     pSrc       points to the Q15 input vector
50   @param[out]    pDst       points to the f16 output vector
51   @param[in]     blockSize  number of samples in each vector
52 
53   @par           Details
54                    The equation used for the conversion process is:
55   <pre>
56       pDst[n] = (float16_t) pSrc[n] / 32768;   0 <= n < blockSize.
57   </pre>
58  */
59 
60 #if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
61 
arm_q15_to_f16(const q15_t * pSrc,float16_t * pDst,uint32_t blockSize)62 ARM_DSP_ATTRIBUTE void arm_q15_to_f16(
63   const q15_t * pSrc,
64   float16_t * pDst,
65   uint32_t blockSize)
66 {
67     int32_t  blkCnt;           /* loop counters */
68     q15x8_t vecDst;
69     q15_t const *pSrcVec;
70 
71     pSrcVec = (q15_t const *) pSrc;
72     blkCnt = blockSize >> 3;
73     while (blkCnt > 0)
74     {
75         /* C = (float16_t) A / 32768 */
76         /* convert from q15 to float and then store the results in the destination buffer */
77         vecDst = vld1q(pSrcVec);   pSrcVec += 8;
78         vstrhq(pDst, vcvtq_n_f16_s16(vecDst, 15));  pDst += 8;
79         /*
80          * Decrement the blockSize loop counter
81          */
82         blkCnt--;
83     }
84     /*
85      * tail
86      * (will be merged thru tail predication)
87      */
88     blkCnt = blockSize & 7;
89     if (blkCnt > 0)
90     {
91         mve_pred16_t p0 = vctp16q(blkCnt);
92         vecDst = vld1q(pSrcVec);   pSrcVec += 8;
93         vstrhq_p(pDst, vcvtq_n_f16_s16(vecDst, 15), p0);
94     }
95 }
96 #else
97 
arm_q15_to_f16(const q15_t * pSrc,float16_t * pDst,uint32_t blockSize)98 ARM_DSP_ATTRIBUTE void arm_q15_to_f16(
99   const q15_t * pSrc,
100         float16_t * pDst,
101         uint32_t blockSize)
102 {
103         uint32_t blkCnt;                               /* Loop counter */
104   const q15_t *pIn = pSrc;                             /* Source pointer */
105 
106 #if defined (ARM_MATH_LOOPUNROLL)
107 
108   /* Loop unrolling: Compute 4 outputs at a time */
109   blkCnt = blockSize >> 2U;
110 
111   while (blkCnt > 0U)
112   {
113     /* C = (float16_t) A / 32768 */
114 
115     /* Convert from q15 to float and store result in destination buffer */
116     *pDst++ = ((_Float16) * pIn++ / 32768.0f16);
117     *pDst++ = ((_Float16) * pIn++ / 32768.0f16);
118     *pDst++ = ((_Float16) * pIn++ / 32768.0f16);
119     *pDst++ = ((_Float16) * pIn++ / 32768.0f16);
120 
121     /* Decrement loop counter */
122     blkCnt--;
123   }
124 
125   /* Loop unrolling: Compute remaining outputs */
126   blkCnt = blockSize % 0x4U;
127 
128 #else
129 
130   /* Initialize blkCnt with number of samples */
131   blkCnt = blockSize;
132 
133 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
134 
135   while (blkCnt > 0U)
136   {
137     /* C = (float16_t) A / 32768 */
138 
139     /* Convert from q15 to float and store result in destination buffer */
140     *pDst++ = ((_Float16) *pIn++ / 32768.0f16);
141 
142     /* Decrement loop counter */
143     blkCnt--;
144   }
145 
146 }
147 #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
148 
149 /**
150   @} end of q15_to_x group
151  */
152 
153 #endif /* #if defined(ARM_FLOAT16_SUPPORTED) */
154 
155