1 /* ----------------------------------------------------------------------
2  * Project:      CMSIS DSP Library
3  * Title:        arm_mat_sub_q31.c
4  * Description:  Q31 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         Q31 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 Q31 range [0x80000000 0x7FFFFFFF] are saturated.
52  */
53 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
arm_mat_sub_q31(const arm_matrix_instance_q31 * pSrcA,const arm_matrix_instance_q31 * pSrcB,arm_matrix_instance_q31 * pDst)54 arm_status arm_mat_sub_q31(
55   const arm_matrix_instance_q31 * pSrcA,
56   const arm_matrix_instance_q31 * pSrcB,
57         arm_matrix_instance_q31 * pDst)
58 {
59     uint32_t        numSamples;       /* total number of elements in the matrix  */
60     q31_t          *pDataA, *pDataB, *pDataDst;
61     q31x4_t       vecA, vecB, vecDst;
62     q31_t const   *pSrcAVec;
63     q31_t const   *pSrcBVec;
64     uint32_t        blkCnt;           /* loop counters */
65     arm_status status;                             /* status of matrix subtraction */
66 
67     pDataA = pSrcA->pData;
68     pDataB = pSrcB->pData;
69     pDataDst = pDst->pData;
70     pSrcAVec = (q31_t const *) pDataA;
71     pSrcBVec = (q31_t const *) pDataB;
72 
73 #ifdef ARM_MATH_MATRIX_CHECK
74 
75   /* Check for matrix mismatch condition */
76   if ((pSrcA->numRows != pSrcB->numRows) ||
77       (pSrcA->numCols != pSrcB->numCols) ||
78       (pSrcA->numRows != pDst->numRows)  ||
79       (pSrcA->numCols != pDst->numCols)    )
80   {
81     /* Set status as ARM_MATH_SIZE_MISMATCH */
82     status = ARM_MATH_SIZE_MISMATCH;
83   }
84   else
85 
86 #endif /* #ifdef ARM_MATH_MATRIX_CHECK */
87   {
88 
89     /*
90      * Total number of samples in the input matrix
91      */
92     numSamples = (uint32_t) pSrcA->numRows * pSrcA->numCols;
93     blkCnt = numSamples >> 2;
94     while (blkCnt > 0U)
95     {
96         /* C(m,n) = A(m,n) + B(m,n) */
97         /* sub and then store the results in the destination buffer. */
98         vecA = vld1q(pSrcAVec);
99         pSrcAVec += 4;
100         vecB = vld1q(pSrcBVec);
101         pSrcBVec += 4;
102         vecDst = vqsubq(vecA, vecB);
103         vst1q(pDataDst, vecDst);
104         pDataDst += 4;
105         /*
106          * Decrement the blockSize loop counter
107          */
108         blkCnt--;
109     }
110     /*
111      * tail
112      */
113     blkCnt = numSamples & 3;
114     if (blkCnt > 0U)
115     {
116         mve_pred16_t p0 = vctp32q(blkCnt);
117         vecA = vld1q(pSrcAVec);
118         pSrcAVec += 4;
119         vecB = vld1q(pSrcBVec);
120         pSrcBVec += 4;
121         vecDst = vqsubq_m(vecDst, vecA, vecB, p0);
122         vstrwq_p(pDataDst, vecDst, p0);
123     }
124     status = ARM_MATH_SUCCESS;
125   }
126 
127   /* Return to application */
128   return (status);
129 }
130 
131 #else
arm_mat_sub_q31(const arm_matrix_instance_q31 * pSrcA,const arm_matrix_instance_q31 * pSrcB,arm_matrix_instance_q31 * pDst)132 arm_status arm_mat_sub_q31(
133   const arm_matrix_instance_q31 * pSrcA,
134   const arm_matrix_instance_q31 * pSrcB,
135         arm_matrix_instance_q31 * pDst)
136 {
137   q31_t *pInA = pSrcA->pData;                    /* input data matrix pointer A */
138   q31_t *pInB = pSrcB->pData;                    /* input data matrix pointer B */
139   q31_t *pOut = pDst->pData;                     /* output data matrix pointer */
140 
141   uint32_t numSamples;                           /* total number of elements in the matrix */
142   uint32_t blkCnt;                               /* loop counters */
143   arm_status status;                             /* status of matrix subtraction */
144 
145 #ifdef ARM_MATH_MATRIX_CHECK
146 
147   /* Check for matrix mismatch condition */
148   if ((pSrcA->numRows != pSrcB->numRows) ||
149       (pSrcA->numCols != pSrcB->numCols) ||
150       (pSrcA->numRows != pDst->numRows)  ||
151       (pSrcA->numCols != pDst->numCols)    )
152   {
153     /* Set status as ARM_MATH_SIZE_MISMATCH */
154     status = ARM_MATH_SIZE_MISMATCH;
155   }
156   else
157 
158 #endif /* #ifdef ARM_MATH_MATRIX_CHECK */
159 
160   {
161     /* Total number of samples in input matrix */
162     numSamples = (uint32_t) pSrcA->numRows * pSrcA->numCols;
163 
164 #if defined (ARM_MATH_LOOPUNROLL)
165 
166     /* Loop unrolling: Compute 4 outputs at a time */
167     blkCnt = numSamples >> 2U;
168 
169     while (blkCnt > 0U)
170     {
171       /* C(m,n) = A(m,n) - B(m,n) */
172 
173       /* Subtract, saturate and then store the results in the destination buffer. */
174       *pOut++ = __QSUB(*pInA++, *pInB++);
175 
176       *pOut++ = __QSUB(*pInA++, *pInB++);
177 
178       *pOut++ = __QSUB(*pInA++, *pInB++);
179 
180       *pOut++ = __QSUB(*pInA++, *pInB++);
181 
182       /* Decrement loop counter */
183       blkCnt--;
184     }
185 
186     /* Loop unrolling: Compute remaining outputs */
187     blkCnt = numSamples % 0x4U;
188 
189 #else
190 
191     /* Initialize blkCnt with number of samples */
192     blkCnt = numSamples;
193 
194 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
195 
196     while (blkCnt > 0U)
197     {
198       /* C(m,n) = A(m,n) - B(m,n) */
199 
200       /* Subtract, saturate and store result in destination buffer. */
201       *pOut++ = __QSUB(*pInA++, *pInB++);
202 
203       /* Decrement loop counter */
204       blkCnt--;
205     }
206 
207     /* Set status as ARM_MATH_SUCCESS */
208     status = ARM_MATH_SUCCESS;
209   }
210 
211   /* Return to application */
212   return (status);
213 }
214 #endif /* defined(ARM_MATH_MVEI) */
215 
216 /**
217   @} end of MatrixSub group
218  */
219