The CMSIS DSP 5 library
Dependents: Nucleo-Heart-Rate ejercicioVrms2 PROYECTOFINAL ejercicioVrms ... more
functions/MatrixFunctions/arm_mat_mult_q15.c@1:24714b45cd1b, 2018-06-20 (annotated)
- Committer:
- xorjoep
- Date:
- Wed Jun 20 11:21:31 2018 +0000
- Revision:
- 1:24714b45cd1b
The newest version of the CMSIS library
Who changed what in which revision?
User | Revision | Line number | New contents of line |
---|---|---|---|
xorjoep | 1:24714b45cd1b | 1 | /* ---------------------------------------------------------------------- |
xorjoep | 1:24714b45cd1b | 2 | * Project: CMSIS DSP Library |
xorjoep | 1:24714b45cd1b | 3 | * Title: arm_mat_mult_q15.c |
xorjoep | 1:24714b45cd1b | 4 | * Description: Q15 matrix multiplication |
xorjoep | 1:24714b45cd1b | 5 | * |
xorjoep | 1:24714b45cd1b | 6 | * $Date: 27. January 2017 |
xorjoep | 1:24714b45cd1b | 7 | * $Revision: V.1.5.1 |
xorjoep | 1:24714b45cd1b | 8 | * |
xorjoep | 1:24714b45cd1b | 9 | * Target Processor: Cortex-M cores |
xorjoep | 1:24714b45cd1b | 10 | * -------------------------------------------------------------------- */ |
xorjoep | 1:24714b45cd1b | 11 | /* |
xorjoep | 1:24714b45cd1b | 12 | * Copyright (C) 2010-2017 ARM Limited or its affiliates. All rights reserved. |
xorjoep | 1:24714b45cd1b | 13 | * |
xorjoep | 1:24714b45cd1b | 14 | * SPDX-License-Identifier: Apache-2.0 |
xorjoep | 1:24714b45cd1b | 15 | * |
xorjoep | 1:24714b45cd1b | 16 | * Licensed under the Apache License, Version 2.0 (the License); you may |
xorjoep | 1:24714b45cd1b | 17 | * not use this file except in compliance with the License. |
xorjoep | 1:24714b45cd1b | 18 | * You may obtain a copy of the License at |
xorjoep | 1:24714b45cd1b | 19 | * |
xorjoep | 1:24714b45cd1b | 20 | * www.apache.org/licenses/LICENSE-2.0 |
xorjoep | 1:24714b45cd1b | 21 | * |
xorjoep | 1:24714b45cd1b | 22 | * Unless required by applicable law or agreed to in writing, software |
xorjoep | 1:24714b45cd1b | 23 | * distributed under the License is distributed on an AS IS BASIS, WITHOUT |
xorjoep | 1:24714b45cd1b | 24 | * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
xorjoep | 1:24714b45cd1b | 25 | * See the License for the specific language governing permissions and |
xorjoep | 1:24714b45cd1b | 26 | * limitations under the License. |
xorjoep | 1:24714b45cd1b | 27 | */ |
xorjoep | 1:24714b45cd1b | 28 | |
xorjoep | 1:24714b45cd1b | 29 | #include "arm_math.h" |
xorjoep | 1:24714b45cd1b | 30 | |
xorjoep | 1:24714b45cd1b | 31 | /** |
xorjoep | 1:24714b45cd1b | 32 | * @ingroup groupMatrix |
xorjoep | 1:24714b45cd1b | 33 | */ |
xorjoep | 1:24714b45cd1b | 34 | |
xorjoep | 1:24714b45cd1b | 35 | /** |
xorjoep | 1:24714b45cd1b | 36 | * @addtogroup MatrixMult |
xorjoep | 1:24714b45cd1b | 37 | * @{ |
xorjoep | 1:24714b45cd1b | 38 | */ |
xorjoep | 1:24714b45cd1b | 39 | |
xorjoep | 1:24714b45cd1b | 40 | |
xorjoep | 1:24714b45cd1b | 41 | /** |
xorjoep | 1:24714b45cd1b | 42 | * @brief Q15 matrix multiplication |
xorjoep | 1:24714b45cd1b | 43 | * @param[in] *pSrcA points to the first input matrix structure |
xorjoep | 1:24714b45cd1b | 44 | * @param[in] *pSrcB points to the second input matrix structure |
xorjoep | 1:24714b45cd1b | 45 | * @param[out] *pDst points to output matrix structure |
xorjoep | 1:24714b45cd1b | 46 | * @param[in] *pState points to the array for storing intermediate results (Unused) |
xorjoep | 1:24714b45cd1b | 47 | * @return The function returns either |
xorjoep | 1:24714b45cd1b | 48 | * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking. |
xorjoep | 1:24714b45cd1b | 49 | * |
xorjoep | 1:24714b45cd1b | 50 | * @details |
xorjoep | 1:24714b45cd1b | 51 | * <b>Scaling and Overflow Behavior:</b> |
xorjoep | 1:24714b45cd1b | 52 | * |
xorjoep | 1:24714b45cd1b | 53 | * \par |
xorjoep | 1:24714b45cd1b | 54 | * The function is implemented using a 64-bit internal accumulator. The inputs to the |
xorjoep | 1:24714b45cd1b | 55 | * multiplications are in 1.15 format and multiplications yield a 2.30 result. |
xorjoep | 1:24714b45cd1b | 56 | * The 2.30 intermediate |
xorjoep | 1:24714b45cd1b | 57 | * results are accumulated in a 64-bit accumulator in 34.30 format. This approach |
xorjoep | 1:24714b45cd1b | 58 | * provides 33 guard bits and there is no risk of overflow. The 34.30 result is then |
xorjoep | 1:24714b45cd1b | 59 | * truncated to 34.15 format by discarding the low 15 bits and then saturated to |
xorjoep | 1:24714b45cd1b | 60 | * 1.15 format. |
xorjoep | 1:24714b45cd1b | 61 | * |
xorjoep | 1:24714b45cd1b | 62 | * \par |
xorjoep | 1:24714b45cd1b | 63 | * Refer to <code>arm_mat_mult_fast_q15()</code> for a faster but less precise version of this function for Cortex-M3 and Cortex-M4. |
xorjoep | 1:24714b45cd1b | 64 | * |
xorjoep | 1:24714b45cd1b | 65 | */ |
xorjoep | 1:24714b45cd1b | 66 | |
xorjoep | 1:24714b45cd1b | 67 | arm_status arm_mat_mult_q15( |
xorjoep | 1:24714b45cd1b | 68 | const arm_matrix_instance_q15 * pSrcA, |
xorjoep | 1:24714b45cd1b | 69 | const arm_matrix_instance_q15 * pSrcB, |
xorjoep | 1:24714b45cd1b | 70 | arm_matrix_instance_q15 * pDst, |
xorjoep | 1:24714b45cd1b | 71 | q15_t * pState) |
xorjoep | 1:24714b45cd1b | 72 | { |
xorjoep | 1:24714b45cd1b | 73 | q63_t sum; /* accumulator */ |
xorjoep | 1:24714b45cd1b | 74 | |
xorjoep | 1:24714b45cd1b | 75 | #if defined (ARM_MATH_DSP) |
xorjoep | 1:24714b45cd1b | 76 | |
xorjoep | 1:24714b45cd1b | 77 | /* Run the below code for Cortex-M4 and Cortex-M3 */ |
xorjoep | 1:24714b45cd1b | 78 | |
xorjoep | 1:24714b45cd1b | 79 | q15_t *pSrcBT = pState; /* input data matrix pointer for transpose */ |
xorjoep | 1:24714b45cd1b | 80 | q15_t *pInA = pSrcA->pData; /* input data matrix pointer A of Q15 type */ |
xorjoep | 1:24714b45cd1b | 81 | q15_t *pInB = pSrcB->pData; /* input data matrix pointer B of Q15 type */ |
xorjoep | 1:24714b45cd1b | 82 | q15_t *px; /* Temporary output data matrix pointer */ |
xorjoep | 1:24714b45cd1b | 83 | uint16_t numRowsA = pSrcA->numRows; /* number of rows of input matrix A */ |
xorjoep | 1:24714b45cd1b | 84 | uint16_t numColsB = pSrcB->numCols; /* number of columns of input matrix B */ |
xorjoep | 1:24714b45cd1b | 85 | uint16_t numColsA = pSrcA->numCols; /* number of columns of input matrix A */ |
xorjoep | 1:24714b45cd1b | 86 | uint16_t numRowsB = pSrcB->numRows; /* number of rows of input matrix A */ |
xorjoep | 1:24714b45cd1b | 87 | uint16_t col, i = 0U, row = numRowsB, colCnt; /* loop counters */ |
xorjoep | 1:24714b45cd1b | 88 | arm_status status; /* status of matrix multiplication */ |
xorjoep | 1:24714b45cd1b | 89 | |
xorjoep | 1:24714b45cd1b | 90 | #ifndef UNALIGNED_SUPPORT_DISABLE |
xorjoep | 1:24714b45cd1b | 91 | |
xorjoep | 1:24714b45cd1b | 92 | q31_t in; /* Temporary variable to hold the input value */ |
xorjoep | 1:24714b45cd1b | 93 | q31_t pSourceA1, pSourceB1, pSourceA2, pSourceB2; |
xorjoep | 1:24714b45cd1b | 94 | |
xorjoep | 1:24714b45cd1b | 95 | #else |
xorjoep | 1:24714b45cd1b | 96 | |
xorjoep | 1:24714b45cd1b | 97 | q15_t in; /* Temporary variable to hold the input value */ |
xorjoep | 1:24714b45cd1b | 98 | q15_t inA1, inB1, inA2, inB2; |
xorjoep | 1:24714b45cd1b | 99 | |
xorjoep | 1:24714b45cd1b | 100 | #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */ |
xorjoep | 1:24714b45cd1b | 101 | |
xorjoep | 1:24714b45cd1b | 102 | #ifdef ARM_MATH_MATRIX_CHECK |
xorjoep | 1:24714b45cd1b | 103 | /* Check for matrix mismatch condition */ |
xorjoep | 1:24714b45cd1b | 104 | if ((pSrcA->numCols != pSrcB->numRows) || |
xorjoep | 1:24714b45cd1b | 105 | (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols)) |
xorjoep | 1:24714b45cd1b | 106 | { |
xorjoep | 1:24714b45cd1b | 107 | /* Set status as ARM_MATH_SIZE_MISMATCH */ |
xorjoep | 1:24714b45cd1b | 108 | status = ARM_MATH_SIZE_MISMATCH; |
xorjoep | 1:24714b45cd1b | 109 | } |
xorjoep | 1:24714b45cd1b | 110 | else |
xorjoep | 1:24714b45cd1b | 111 | #endif /* #ifdef ARM_MATH_MATRIX_CHECK */ |
xorjoep | 1:24714b45cd1b | 112 | { |
xorjoep | 1:24714b45cd1b | 113 | /* Matrix transpose */ |
xorjoep | 1:24714b45cd1b | 114 | do |
xorjoep | 1:24714b45cd1b | 115 | { |
xorjoep | 1:24714b45cd1b | 116 | /* Apply loop unrolling and exchange the columns with row elements */ |
xorjoep | 1:24714b45cd1b | 117 | col = numColsB >> 2; |
xorjoep | 1:24714b45cd1b | 118 | |
xorjoep | 1:24714b45cd1b | 119 | /* The pointer px is set to starting address of the column being processed */ |
xorjoep | 1:24714b45cd1b | 120 | px = pSrcBT + i; |
xorjoep | 1:24714b45cd1b | 121 | |
xorjoep | 1:24714b45cd1b | 122 | /* First part of the processing with loop unrolling. Compute 4 outputs at a time. |
xorjoep | 1:24714b45cd1b | 123 | ** a second loop below computes the remaining 1 to 3 samples. */ |
xorjoep | 1:24714b45cd1b | 124 | while (col > 0U) |
xorjoep | 1:24714b45cd1b | 125 | { |
xorjoep | 1:24714b45cd1b | 126 | #ifndef UNALIGNED_SUPPORT_DISABLE |
xorjoep | 1:24714b45cd1b | 127 | |
xorjoep | 1:24714b45cd1b | 128 | /* Read two elements from the row */ |
xorjoep | 1:24714b45cd1b | 129 | in = *__SIMD32(pInB)++; |
xorjoep | 1:24714b45cd1b | 130 | |
xorjoep | 1:24714b45cd1b | 131 | /* Unpack and store one element in the destination */ |
xorjoep | 1:24714b45cd1b | 132 | #ifndef ARM_MATH_BIG_ENDIAN |
xorjoep | 1:24714b45cd1b | 133 | |
xorjoep | 1:24714b45cd1b | 134 | *px = (q15_t) in; |
xorjoep | 1:24714b45cd1b | 135 | |
xorjoep | 1:24714b45cd1b | 136 | #else |
xorjoep | 1:24714b45cd1b | 137 | |
xorjoep | 1:24714b45cd1b | 138 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
xorjoep | 1:24714b45cd1b | 139 | |
xorjoep | 1:24714b45cd1b | 140 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
xorjoep | 1:24714b45cd1b | 141 | |
xorjoep | 1:24714b45cd1b | 142 | /* Update the pointer px to point to the next row of the transposed matrix */ |
xorjoep | 1:24714b45cd1b | 143 | px += numRowsB; |
xorjoep | 1:24714b45cd1b | 144 | |
xorjoep | 1:24714b45cd1b | 145 | /* Unpack and store the second element in the destination */ |
xorjoep | 1:24714b45cd1b | 146 | #ifndef ARM_MATH_BIG_ENDIAN |
xorjoep | 1:24714b45cd1b | 147 | |
xorjoep | 1:24714b45cd1b | 148 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
xorjoep | 1:24714b45cd1b | 149 | |
xorjoep | 1:24714b45cd1b | 150 | #else |
xorjoep | 1:24714b45cd1b | 151 | |
xorjoep | 1:24714b45cd1b | 152 | *px = (q15_t) in; |
xorjoep | 1:24714b45cd1b | 153 | |
xorjoep | 1:24714b45cd1b | 154 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
xorjoep | 1:24714b45cd1b | 155 | |
xorjoep | 1:24714b45cd1b | 156 | /* Update the pointer px to point to the next row of the transposed matrix */ |
xorjoep | 1:24714b45cd1b | 157 | px += numRowsB; |
xorjoep | 1:24714b45cd1b | 158 | |
xorjoep | 1:24714b45cd1b | 159 | /* Read two elements from the row */ |
xorjoep | 1:24714b45cd1b | 160 | in = *__SIMD32(pInB)++; |
xorjoep | 1:24714b45cd1b | 161 | |
xorjoep | 1:24714b45cd1b | 162 | /* Unpack and store one element in the destination */ |
xorjoep | 1:24714b45cd1b | 163 | #ifndef ARM_MATH_BIG_ENDIAN |
xorjoep | 1:24714b45cd1b | 164 | |
xorjoep | 1:24714b45cd1b | 165 | *px = (q15_t) in; |
xorjoep | 1:24714b45cd1b | 166 | |
xorjoep | 1:24714b45cd1b | 167 | #else |
xorjoep | 1:24714b45cd1b | 168 | |
xorjoep | 1:24714b45cd1b | 169 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
xorjoep | 1:24714b45cd1b | 170 | |
xorjoep | 1:24714b45cd1b | 171 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
xorjoep | 1:24714b45cd1b | 172 | |
xorjoep | 1:24714b45cd1b | 173 | /* Update the pointer px to point to the next row of the transposed matrix */ |
xorjoep | 1:24714b45cd1b | 174 | px += numRowsB; |
xorjoep | 1:24714b45cd1b | 175 | |
xorjoep | 1:24714b45cd1b | 176 | /* Unpack and store the second element in the destination */ |
xorjoep | 1:24714b45cd1b | 177 | |
xorjoep | 1:24714b45cd1b | 178 | #ifndef ARM_MATH_BIG_ENDIAN |
xorjoep | 1:24714b45cd1b | 179 | |
xorjoep | 1:24714b45cd1b | 180 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
xorjoep | 1:24714b45cd1b | 181 | |
xorjoep | 1:24714b45cd1b | 182 | #else |
xorjoep | 1:24714b45cd1b | 183 | |
xorjoep | 1:24714b45cd1b | 184 | *px = (q15_t) in; |
xorjoep | 1:24714b45cd1b | 185 | |
xorjoep | 1:24714b45cd1b | 186 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
xorjoep | 1:24714b45cd1b | 187 | |
xorjoep | 1:24714b45cd1b | 188 | /* Update the pointer px to point to the next row of the transposed matrix */ |
xorjoep | 1:24714b45cd1b | 189 | px += numRowsB; |
xorjoep | 1:24714b45cd1b | 190 | |
xorjoep | 1:24714b45cd1b | 191 | #else |
xorjoep | 1:24714b45cd1b | 192 | |
xorjoep | 1:24714b45cd1b | 193 | /* Read one element from the row */ |
xorjoep | 1:24714b45cd1b | 194 | in = *pInB++; |
xorjoep | 1:24714b45cd1b | 195 | |
xorjoep | 1:24714b45cd1b | 196 | /* Store one element in the destination */ |
xorjoep | 1:24714b45cd1b | 197 | *px = in; |
xorjoep | 1:24714b45cd1b | 198 | |
xorjoep | 1:24714b45cd1b | 199 | /* Update the pointer px to point to the next row of the transposed matrix */ |
xorjoep | 1:24714b45cd1b | 200 | px += numRowsB; |
xorjoep | 1:24714b45cd1b | 201 | |
xorjoep | 1:24714b45cd1b | 202 | /* Read one element from the row */ |
xorjoep | 1:24714b45cd1b | 203 | in = *pInB++; |
xorjoep | 1:24714b45cd1b | 204 | |
xorjoep | 1:24714b45cd1b | 205 | /* Store one element in the destination */ |
xorjoep | 1:24714b45cd1b | 206 | *px = in; |
xorjoep | 1:24714b45cd1b | 207 | |
xorjoep | 1:24714b45cd1b | 208 | /* Update the pointer px to point to the next row of the transposed matrix */ |
xorjoep | 1:24714b45cd1b | 209 | px += numRowsB; |
xorjoep | 1:24714b45cd1b | 210 | |
xorjoep | 1:24714b45cd1b | 211 | /* Read one element from the row */ |
xorjoep | 1:24714b45cd1b | 212 | in = *pInB++; |
xorjoep | 1:24714b45cd1b | 213 | |
xorjoep | 1:24714b45cd1b | 214 | /* Store one element in the destination */ |
xorjoep | 1:24714b45cd1b | 215 | *px = in; |
xorjoep | 1:24714b45cd1b | 216 | |
xorjoep | 1:24714b45cd1b | 217 | /* Update the pointer px to point to the next row of the transposed matrix */ |
xorjoep | 1:24714b45cd1b | 218 | px += numRowsB; |
xorjoep | 1:24714b45cd1b | 219 | |
xorjoep | 1:24714b45cd1b | 220 | /* Read one element from the row */ |
xorjoep | 1:24714b45cd1b | 221 | in = *pInB++; |
xorjoep | 1:24714b45cd1b | 222 | |
xorjoep | 1:24714b45cd1b | 223 | /* Store one element in the destination */ |
xorjoep | 1:24714b45cd1b | 224 | *px = in; |
xorjoep | 1:24714b45cd1b | 225 | |
xorjoep | 1:24714b45cd1b | 226 | /* Update the pointer px to point to the next row of the transposed matrix */ |
xorjoep | 1:24714b45cd1b | 227 | px += numRowsB; |
xorjoep | 1:24714b45cd1b | 228 | |
xorjoep | 1:24714b45cd1b | 229 | #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */ |
xorjoep | 1:24714b45cd1b | 230 | |
xorjoep | 1:24714b45cd1b | 231 | /* Decrement the column loop counter */ |
xorjoep | 1:24714b45cd1b | 232 | col--; |
xorjoep | 1:24714b45cd1b | 233 | } |
xorjoep | 1:24714b45cd1b | 234 | |
xorjoep | 1:24714b45cd1b | 235 | /* If the columns of pSrcB is not a multiple of 4, compute any remaining output samples here. |
xorjoep | 1:24714b45cd1b | 236 | ** No loop unrolling is used. */ |
xorjoep | 1:24714b45cd1b | 237 | col = numColsB % 0x4U; |
xorjoep | 1:24714b45cd1b | 238 | |
xorjoep | 1:24714b45cd1b | 239 | while (col > 0U) |
xorjoep | 1:24714b45cd1b | 240 | { |
xorjoep | 1:24714b45cd1b | 241 | /* Read and store the input element in the destination */ |
xorjoep | 1:24714b45cd1b | 242 | *px = *pInB++; |
xorjoep | 1:24714b45cd1b | 243 | |
xorjoep | 1:24714b45cd1b | 244 | /* Update the pointer px to point to the next row of the transposed matrix */ |
xorjoep | 1:24714b45cd1b | 245 | px += numRowsB; |
xorjoep | 1:24714b45cd1b | 246 | |
xorjoep | 1:24714b45cd1b | 247 | /* Decrement the column loop counter */ |
xorjoep | 1:24714b45cd1b | 248 | col--; |
xorjoep | 1:24714b45cd1b | 249 | } |
xorjoep | 1:24714b45cd1b | 250 | |
xorjoep | 1:24714b45cd1b | 251 | i++; |
xorjoep | 1:24714b45cd1b | 252 | |
xorjoep | 1:24714b45cd1b | 253 | /* Decrement the row loop counter */ |
xorjoep | 1:24714b45cd1b | 254 | row--; |
xorjoep | 1:24714b45cd1b | 255 | |
xorjoep | 1:24714b45cd1b | 256 | } while (row > 0U); |
xorjoep | 1:24714b45cd1b | 257 | |
xorjoep | 1:24714b45cd1b | 258 | /* Reset the variables for the usage in the following multiplication process */ |
xorjoep | 1:24714b45cd1b | 259 | row = numRowsA; |
xorjoep | 1:24714b45cd1b | 260 | i = 0U; |
xorjoep | 1:24714b45cd1b | 261 | px = pDst->pData; |
xorjoep | 1:24714b45cd1b | 262 | |
xorjoep | 1:24714b45cd1b | 263 | /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */ |
xorjoep | 1:24714b45cd1b | 264 | /* row loop */ |
xorjoep | 1:24714b45cd1b | 265 | do |
xorjoep | 1:24714b45cd1b | 266 | { |
xorjoep | 1:24714b45cd1b | 267 | /* For every row wise process, the column loop counter is to be initiated */ |
xorjoep | 1:24714b45cd1b | 268 | col = numColsB; |
xorjoep | 1:24714b45cd1b | 269 | |
xorjoep | 1:24714b45cd1b | 270 | /* For every row wise process, the pIn2 pointer is set |
xorjoep | 1:24714b45cd1b | 271 | ** to the starting address of the transposed pSrcB data */ |
xorjoep | 1:24714b45cd1b | 272 | pInB = pSrcBT; |
xorjoep | 1:24714b45cd1b | 273 | |
xorjoep | 1:24714b45cd1b | 274 | /* column loop */ |
xorjoep | 1:24714b45cd1b | 275 | do |
xorjoep | 1:24714b45cd1b | 276 | { |
xorjoep | 1:24714b45cd1b | 277 | /* Set the variable sum, that acts as accumulator, to zero */ |
xorjoep | 1:24714b45cd1b | 278 | sum = 0; |
xorjoep | 1:24714b45cd1b | 279 | |
xorjoep | 1:24714b45cd1b | 280 | /* Apply loop unrolling and compute 2 MACs simultaneously. */ |
xorjoep | 1:24714b45cd1b | 281 | colCnt = numColsA >> 2; |
xorjoep | 1:24714b45cd1b | 282 | |
xorjoep | 1:24714b45cd1b | 283 | /* Initiate the pointer pIn1 to point to the starting address of the column being processed */ |
xorjoep | 1:24714b45cd1b | 284 | pInA = pSrcA->pData + i; |
xorjoep | 1:24714b45cd1b | 285 | |
xorjoep | 1:24714b45cd1b | 286 | |
xorjoep | 1:24714b45cd1b | 287 | /* matrix multiplication */ |
xorjoep | 1:24714b45cd1b | 288 | while (colCnt > 0U) |
xorjoep | 1:24714b45cd1b | 289 | { |
xorjoep | 1:24714b45cd1b | 290 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
xorjoep | 1:24714b45cd1b | 291 | #ifndef UNALIGNED_SUPPORT_DISABLE |
xorjoep | 1:24714b45cd1b | 292 | |
xorjoep | 1:24714b45cd1b | 293 | /* read real and imag values from pSrcA and pSrcB buffer */ |
xorjoep | 1:24714b45cd1b | 294 | pSourceA1 = *__SIMD32(pInA)++; |
xorjoep | 1:24714b45cd1b | 295 | pSourceB1 = *__SIMD32(pInB)++; |
xorjoep | 1:24714b45cd1b | 296 | |
xorjoep | 1:24714b45cd1b | 297 | pSourceA2 = *__SIMD32(pInA)++; |
xorjoep | 1:24714b45cd1b | 298 | pSourceB2 = *__SIMD32(pInB)++; |
xorjoep | 1:24714b45cd1b | 299 | |
xorjoep | 1:24714b45cd1b | 300 | /* Multiply and Accumlates */ |
xorjoep | 1:24714b45cd1b | 301 | sum = __SMLALD(pSourceA1, pSourceB1, sum); |
xorjoep | 1:24714b45cd1b | 302 | sum = __SMLALD(pSourceA2, pSourceB2, sum); |
xorjoep | 1:24714b45cd1b | 303 | |
xorjoep | 1:24714b45cd1b | 304 | #else |
xorjoep | 1:24714b45cd1b | 305 | /* read real and imag values from pSrcA and pSrcB buffer */ |
xorjoep | 1:24714b45cd1b | 306 | inA1 = *pInA++; |
xorjoep | 1:24714b45cd1b | 307 | inB1 = *pInB++; |
xorjoep | 1:24714b45cd1b | 308 | inA2 = *pInA++; |
xorjoep | 1:24714b45cd1b | 309 | /* Multiply and Accumlates */ |
xorjoep | 1:24714b45cd1b | 310 | sum += inA1 * inB1; |
xorjoep | 1:24714b45cd1b | 311 | inB2 = *pInB++; |
xorjoep | 1:24714b45cd1b | 312 | |
xorjoep | 1:24714b45cd1b | 313 | inA1 = *pInA++; |
xorjoep | 1:24714b45cd1b | 314 | inB1 = *pInB++; |
xorjoep | 1:24714b45cd1b | 315 | /* Multiply and Accumlates */ |
xorjoep | 1:24714b45cd1b | 316 | sum += inA2 * inB2; |
xorjoep | 1:24714b45cd1b | 317 | inA2 = *pInA++; |
xorjoep | 1:24714b45cd1b | 318 | inB2 = *pInB++; |
xorjoep | 1:24714b45cd1b | 319 | |
xorjoep | 1:24714b45cd1b | 320 | /* Multiply and Accumlates */ |
xorjoep | 1:24714b45cd1b | 321 | sum += inA1 * inB1; |
xorjoep | 1:24714b45cd1b | 322 | sum += inA2 * inB2; |
xorjoep | 1:24714b45cd1b | 323 | |
xorjoep | 1:24714b45cd1b | 324 | #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */ |
xorjoep | 1:24714b45cd1b | 325 | |
xorjoep | 1:24714b45cd1b | 326 | /* Decrement the loop counter */ |
xorjoep | 1:24714b45cd1b | 327 | colCnt--; |
xorjoep | 1:24714b45cd1b | 328 | } |
xorjoep | 1:24714b45cd1b | 329 | |
xorjoep | 1:24714b45cd1b | 330 | /* process remaining column samples */ |
xorjoep | 1:24714b45cd1b | 331 | colCnt = numColsA & 3U; |
xorjoep | 1:24714b45cd1b | 332 | |
xorjoep | 1:24714b45cd1b | 333 | while (colCnt > 0U) |
xorjoep | 1:24714b45cd1b | 334 | { |
xorjoep | 1:24714b45cd1b | 335 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
xorjoep | 1:24714b45cd1b | 336 | sum += *pInA++ * *pInB++; |
xorjoep | 1:24714b45cd1b | 337 | |
xorjoep | 1:24714b45cd1b | 338 | /* Decrement the loop counter */ |
xorjoep | 1:24714b45cd1b | 339 | colCnt--; |
xorjoep | 1:24714b45cd1b | 340 | } |
xorjoep | 1:24714b45cd1b | 341 | |
xorjoep | 1:24714b45cd1b | 342 | /* Saturate and store the result in the destination buffer */ |
xorjoep | 1:24714b45cd1b | 343 | *px = (q15_t) (__SSAT((sum >> 15), 16)); |
xorjoep | 1:24714b45cd1b | 344 | px++; |
xorjoep | 1:24714b45cd1b | 345 | |
xorjoep | 1:24714b45cd1b | 346 | /* Decrement the column loop counter */ |
xorjoep | 1:24714b45cd1b | 347 | col--; |
xorjoep | 1:24714b45cd1b | 348 | |
xorjoep | 1:24714b45cd1b | 349 | } while (col > 0U); |
xorjoep | 1:24714b45cd1b | 350 | |
xorjoep | 1:24714b45cd1b | 351 | i = i + numColsA; |
xorjoep | 1:24714b45cd1b | 352 | |
xorjoep | 1:24714b45cd1b | 353 | /* Decrement the row loop counter */ |
xorjoep | 1:24714b45cd1b | 354 | row--; |
xorjoep | 1:24714b45cd1b | 355 | |
xorjoep | 1:24714b45cd1b | 356 | } while (row > 0U); |
xorjoep | 1:24714b45cd1b | 357 | |
xorjoep | 1:24714b45cd1b | 358 | #else |
xorjoep | 1:24714b45cd1b | 359 | |
xorjoep | 1:24714b45cd1b | 360 | /* Run the below code for Cortex-M0 */ |
xorjoep | 1:24714b45cd1b | 361 | |
xorjoep | 1:24714b45cd1b | 362 | q15_t *pIn1 = pSrcA->pData; /* input data matrix pointer A */ |
xorjoep | 1:24714b45cd1b | 363 | q15_t *pIn2 = pSrcB->pData; /* input data matrix pointer B */ |
xorjoep | 1:24714b45cd1b | 364 | q15_t *pInA = pSrcA->pData; /* input data matrix pointer A of Q15 type */ |
xorjoep | 1:24714b45cd1b | 365 | q15_t *pInB = pSrcB->pData; /* input data matrix pointer B of Q15 type */ |
xorjoep | 1:24714b45cd1b | 366 | q15_t *pOut = pDst->pData; /* output data matrix pointer */ |
xorjoep | 1:24714b45cd1b | 367 | q15_t *px; /* Temporary output data matrix pointer */ |
xorjoep | 1:24714b45cd1b | 368 | uint16_t numColsB = pSrcB->numCols; /* number of columns of input matrix B */ |
xorjoep | 1:24714b45cd1b | 369 | uint16_t numColsA = pSrcA->numCols; /* number of columns of input matrix A */ |
xorjoep | 1:24714b45cd1b | 370 | uint16_t numRowsA = pSrcA->numRows; /* number of rows of input matrix A */ |
xorjoep | 1:24714b45cd1b | 371 | uint16_t col, i = 0U, row = numRowsA, colCnt; /* loop counters */ |
xorjoep | 1:24714b45cd1b | 372 | arm_status status; /* status of matrix multiplication */ |
xorjoep | 1:24714b45cd1b | 373 | |
xorjoep | 1:24714b45cd1b | 374 | #ifdef ARM_MATH_MATRIX_CHECK |
xorjoep | 1:24714b45cd1b | 375 | |
xorjoep | 1:24714b45cd1b | 376 | /* Check for matrix mismatch condition */ |
xorjoep | 1:24714b45cd1b | 377 | if ((pSrcA->numCols != pSrcB->numRows) || |
xorjoep | 1:24714b45cd1b | 378 | (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols)) |
xorjoep | 1:24714b45cd1b | 379 | { |
xorjoep | 1:24714b45cd1b | 380 | /* Set status as ARM_MATH_SIZE_MISMATCH */ |
xorjoep | 1:24714b45cd1b | 381 | status = ARM_MATH_SIZE_MISMATCH; |
xorjoep | 1:24714b45cd1b | 382 | } |
xorjoep | 1:24714b45cd1b | 383 | else |
xorjoep | 1:24714b45cd1b | 384 | #endif /* #ifdef ARM_MATH_MATRIX_CHECK */ |
xorjoep | 1:24714b45cd1b | 385 | |
xorjoep | 1:24714b45cd1b | 386 | { |
xorjoep | 1:24714b45cd1b | 387 | /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */ |
xorjoep | 1:24714b45cd1b | 388 | /* row loop */ |
xorjoep | 1:24714b45cd1b | 389 | do |
xorjoep | 1:24714b45cd1b | 390 | { |
xorjoep | 1:24714b45cd1b | 391 | /* Output pointer is set to starting address of the row being processed */ |
xorjoep | 1:24714b45cd1b | 392 | px = pOut + i; |
xorjoep | 1:24714b45cd1b | 393 | |
xorjoep | 1:24714b45cd1b | 394 | /* For every row wise process, the column loop counter is to be initiated */ |
xorjoep | 1:24714b45cd1b | 395 | col = numColsB; |
xorjoep | 1:24714b45cd1b | 396 | |
xorjoep | 1:24714b45cd1b | 397 | /* For every row wise process, the pIn2 pointer is set |
xorjoep | 1:24714b45cd1b | 398 | ** to the starting address of the pSrcB data */ |
xorjoep | 1:24714b45cd1b | 399 | pIn2 = pSrcB->pData; |
xorjoep | 1:24714b45cd1b | 400 | |
xorjoep | 1:24714b45cd1b | 401 | /* column loop */ |
xorjoep | 1:24714b45cd1b | 402 | do |
xorjoep | 1:24714b45cd1b | 403 | { |
xorjoep | 1:24714b45cd1b | 404 | /* Set the variable sum, that acts as accumulator, to zero */ |
xorjoep | 1:24714b45cd1b | 405 | sum = 0; |
xorjoep | 1:24714b45cd1b | 406 | |
xorjoep | 1:24714b45cd1b | 407 | /* Initiate the pointer pIn1 to point to the starting address of pSrcA */ |
xorjoep | 1:24714b45cd1b | 408 | pIn1 = pInA; |
xorjoep | 1:24714b45cd1b | 409 | |
xorjoep | 1:24714b45cd1b | 410 | /* Matrix A columns number of MAC operations are to be performed */ |
xorjoep | 1:24714b45cd1b | 411 | colCnt = numColsA; |
xorjoep | 1:24714b45cd1b | 412 | |
xorjoep | 1:24714b45cd1b | 413 | /* matrix multiplication */ |
xorjoep | 1:24714b45cd1b | 414 | while (colCnt > 0U) |
xorjoep | 1:24714b45cd1b | 415 | { |
xorjoep | 1:24714b45cd1b | 416 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
xorjoep | 1:24714b45cd1b | 417 | /* Perform the multiply-accumulates */ |
xorjoep | 1:24714b45cd1b | 418 | sum += (q31_t) * pIn1++ * *pIn2; |
xorjoep | 1:24714b45cd1b | 419 | pIn2 += numColsB; |
xorjoep | 1:24714b45cd1b | 420 | |
xorjoep | 1:24714b45cd1b | 421 | /* Decrement the loop counter */ |
xorjoep | 1:24714b45cd1b | 422 | colCnt--; |
xorjoep | 1:24714b45cd1b | 423 | } |
xorjoep | 1:24714b45cd1b | 424 | |
xorjoep | 1:24714b45cd1b | 425 | /* Convert the result from 34.30 to 1.15 format and store the saturated value in destination buffer */ |
xorjoep | 1:24714b45cd1b | 426 | /* Saturate and store the result in the destination buffer */ |
xorjoep | 1:24714b45cd1b | 427 | *px++ = (q15_t) __SSAT((sum >> 15), 16); |
xorjoep | 1:24714b45cd1b | 428 | |
xorjoep | 1:24714b45cd1b | 429 | /* Decrement the column loop counter */ |
xorjoep | 1:24714b45cd1b | 430 | col--; |
xorjoep | 1:24714b45cd1b | 431 | |
xorjoep | 1:24714b45cd1b | 432 | /* Update the pointer pIn2 to point to the starting address of the next column */ |
xorjoep | 1:24714b45cd1b | 433 | pIn2 = pInB + (numColsB - col); |
xorjoep | 1:24714b45cd1b | 434 | |
xorjoep | 1:24714b45cd1b | 435 | } while (col > 0U); |
xorjoep | 1:24714b45cd1b | 436 | |
xorjoep | 1:24714b45cd1b | 437 | /* Update the pointer pSrcA to point to the starting address of the next row */ |
xorjoep | 1:24714b45cd1b | 438 | i = i + numColsB; |
xorjoep | 1:24714b45cd1b | 439 | pInA = pInA + numColsA; |
xorjoep | 1:24714b45cd1b | 440 | |
xorjoep | 1:24714b45cd1b | 441 | /* Decrement the row loop counter */ |
xorjoep | 1:24714b45cd1b | 442 | row--; |
xorjoep | 1:24714b45cd1b | 443 | |
xorjoep | 1:24714b45cd1b | 444 | } while (row > 0U); |
xorjoep | 1:24714b45cd1b | 445 | |
xorjoep | 1:24714b45cd1b | 446 | #endif /* #if defined (ARM_MATH_DSP) */ |
xorjoep | 1:24714b45cd1b | 447 | /* set status as ARM_MATH_SUCCESS */ |
xorjoep | 1:24714b45cd1b | 448 | status = ARM_MATH_SUCCESS; |
xorjoep | 1:24714b45cd1b | 449 | } |
xorjoep | 1:24714b45cd1b | 450 | |
xorjoep | 1:24714b45cd1b | 451 | /* Return to application */ |
xorjoep | 1:24714b45cd1b | 452 | return (status); |
xorjoep | 1:24714b45cd1b | 453 | } |
xorjoep | 1:24714b45cd1b | 454 | |
xorjoep | 1:24714b45cd1b | 455 | /** |
xorjoep | 1:24714b45cd1b | 456 | * @} end of MatrixMult group |
xorjoep | 1:24714b45cd1b | 457 | */ |