1 /* ----------------------------------------------------------------------
2  * Project:      CMSIS DSP Library
3  * Title:        arm_mat_sub_q15.c
4  * Description:  Q15 Matrix 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/matrix_functions.h"
30 
31 /**
32   @ingroup groupMatrix
33  */
34 
35 /**
36   @addtogroup MatrixSub
37   @{
38  */
39 
40 /**
41   @brief         Q15 matrix subtraction.
42   @param[in]     pSrcA      points to the first input matrix structure
43   @param[in]     pSrcB      points to the second input matrix structure
44   @param[out]    pDst       points to output matrix structure
45   @return        execution status
46                    - \ref ARM_MATH_SUCCESS       : Operation successful
47                    - \ref ARM_MATH_SIZE_MISMATCH : Matrix size check failed
48 
49   @par           Scaling and Overflow Behavior
50                    The function uses saturating arithmetic.
51                    Results outside of the allowable Q15 range [0x8000 0x7FFF] are saturated.
52  */
53 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
54 
arm_mat_sub_q15(const arm_matrix_instance_q15 * pSrcA,const arm_matrix_instance_q15 * pSrcB,arm_matrix_instance_q15 * pDst)55 arm_status arm_mat_sub_q15(
56   const arm_matrix_instance_q15 * pSrcA,
57   const arm_matrix_instance_q15 * pSrcB,
58         arm_matrix_instance_q15 * pDst)
59 {
60     uint32_t        numSamples;       /* total number of elements in the matrix  */
61     q15_t          *pDataA, *pDataB, *pDataDst;
62     q15x8_t       vecA, vecB, vecDst;
63     q15_t const   *pSrcAVec;
64     q15_t const   *pSrcBVec;
65     uint32_t        blkCnt;           /* loop counters */
66     arm_status status;                             /* status of matrix subtraction  */
67 
68 
69     pDataA = pSrcA->pData;
70     pDataB = pSrcB->pData;
71     pDataDst = pDst->pData;
72     pSrcAVec = (q15_t const *) pDataA;
73     pSrcBVec = (q15_t const *) pDataB;
74 
75   #ifdef ARM_MATH_MATRIX_CHECK
76 
77   /* Check for matrix mismatch condition */
78   if ((pSrcA->numRows != pSrcB->numRows) ||
79       (pSrcA->numCols != pSrcB->numCols) ||
80       (pSrcA->numRows != pDst->numRows)  ||
81       (pSrcA->numCols != pDst->numCols)    )
82   {
83     /* Set status as ARM_MATH_SIZE_MISMATCH */
84     status = ARM_MATH_SIZE_MISMATCH;
85   }
86   else
87 #endif /* #ifdef ARM_MATH_MATRIX_CHECK */
88 
89   {
90         /*
91      * Total number of samples in the input matrix
92      */
93     numSamples = (uint32_t) pSrcA->numRows * pSrcA->numCols;
94     blkCnt = numSamples >> 3;
95     while (blkCnt > 0U)
96     {
97         /* C(m,n) = A(m,n) + B(m,n) */
98         /* sub and then store the results in the destination buffer. */
99         vecA = vld1q(pSrcAVec); pSrcAVec += 8;
100         vecB = vld1q(pSrcBVec); pSrcBVec += 8;
101         vecDst = vqsubq(vecA, vecB);
102         vst1q(pDataDst, vecDst);  pDataDst += 8;
103         /*
104          * Decrement the blockSize loop counter
105          */
106         blkCnt--;
107     }
108     /*
109      * tail
110      */
111     blkCnt = numSamples & 7;
112     if (blkCnt > 0U)
113     {
114         mve_pred16_t p0 = vctp16q(blkCnt);
115         vecA = vld1q(pSrcAVec); pSrcAVec += 8;
116         vecB = vld1q(pSrcBVec); pSrcBVec += 8;
117         vecDst = vqsubq_m(vecDst, vecA, vecB, p0);
118         vstrhq_p(pDataDst, vecDst, p0);
119     }
120      status = ARM_MATH_SUCCESS;
121   }
122 
123   /* Return to application */
124   return (status);
125 }
126 
127 #else
arm_mat_sub_q15(const arm_matrix_instance_q15 * pSrcA,const arm_matrix_instance_q15 * pSrcB,arm_matrix_instance_q15 * pDst)128 arm_status arm_mat_sub_q15(
129   const arm_matrix_instance_q15 * pSrcA,
130   const arm_matrix_instance_q15 * pSrcB,
131         arm_matrix_instance_q15 * pDst)
132 {
133         q15_t *pInA = pSrcA->pData;                    /* input data matrix pointer A */
134         q15_t *pInB = pSrcB->pData;                    /* input data matrix pointer B */
135         q15_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 subtraction  */
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 #endif /* #ifdef ARM_MATH_MATRIX_CHECK */
154 
155   {
156     /* Total number of samples in input matrix */
157     numSamples = (uint32_t) pSrcA->numRows * pSrcA->numCols;
158 
159 #if defined (ARM_MATH_LOOPUNROLL)
160 
161     /* Loop unrolling: Compute 4 outputs at a time */
162     blkCnt = numSamples >> 2U;
163 
164     while (blkCnt > 0U)
165     {
166       /* C(m,n) = A(m,n) - B(m,n) */
167 
168       /* Subtract, Saturate and store result in destination buffer. */
169 #if defined (ARM_MATH_DSP)
170       write_q15x2_ia (&pOut, __QSUB16(read_q15x2_ia ((q15_t **) &pInA), read_q15x2_ia ((q15_t **) &pInB)));
171       write_q15x2_ia (&pOut, __QSUB16(read_q15x2_ia ((q15_t **) &pInA), read_q15x2_ia ((q15_t **) &pInB)));
172 #else
173       *pOut++ = (q15_t) __SSAT(((q31_t) * pInA++ - *pInB++), 16);
174       *pOut++ = (q15_t) __SSAT(((q31_t) * pInA++ - *pInB++), 16);
175       *pOut++ = (q15_t) __SSAT(((q31_t) * pInA++ - *pInB++), 16);
176       *pOut++ = (q15_t) __SSAT(((q31_t) * pInA++ - *pInB++), 16);
177 #endif
178 
179       /* Decrement loop counter */
180       blkCnt--;
181     }
182 
183     /* Loop unrolling: Compute remaining outputs */
184     blkCnt = numSamples % 0x4U;
185 
186 #else
187 
188     /* Initialize blkCnt with number of samples */
189     blkCnt = numSamples;
190 
191 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
192 
193     while (blkCnt > 0U)
194     {
195       /* C(m,n) = A(m,n) - B(m,n) */
196 
197       /* Subtract and store result in destination buffer. */
198 #if defined (ARM_MATH_DSP)
199       *pOut++ = (q15_t) __QSUB16(*pInA++, *pInB++);
200 #else
201       *pOut++ = (q15_t) __SSAT(((q31_t) * pInA++ - *pInB++), 16);
202 #endif
203 
204       /* Decrement loop counter */
205       blkCnt--;
206     }
207 
208     /* Set status as ARM_MATH_SUCCESS */
209     status = ARM_MATH_SUCCESS;
210   }
211 
212   /* Return to application */
213   return (status);
214 }
215 #endif /* defined(ARM_MATH_MVEI) */
216 
217 /**
218   @} end of MatrixSub group
219  */
220