The CMSIS DSP 5 library
Dependents: Nucleo-Heart-Rate ejercicioVrms2 PROYECTOFINAL ejercicioVrms ... more
functions/MatrixFunctions/arm_mat_cmplx_mult_f32.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_cmplx_mult_f32.c |
xorjoep | 1:24714b45cd1b | 4 | * Description: Floating-point 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 | * @defgroup CmplxMatrixMult Complex Matrix Multiplication |
xorjoep | 1:24714b45cd1b | 37 | * |
xorjoep | 1:24714b45cd1b | 38 | * Complex Matrix multiplication is only defined if the number of columns of the |
xorjoep | 1:24714b45cd1b | 39 | * first matrix equals the number of rows of the second matrix. |
xorjoep | 1:24714b45cd1b | 40 | * Multiplying an <code>M x N</code> matrix with an <code>N x P</code> matrix results |
xorjoep | 1:24714b45cd1b | 41 | * in an <code>M x P</code> matrix. |
xorjoep | 1:24714b45cd1b | 42 | * When matrix size checking is enabled, the functions check: (1) that the inner dimensions of |
xorjoep | 1:24714b45cd1b | 43 | * <code>pSrcA</code> and <code>pSrcB</code> are equal; and (2) that the size of the output |
xorjoep | 1:24714b45cd1b | 44 | * matrix equals the outer dimensions of <code>pSrcA</code> and <code>pSrcB</code>. |
xorjoep | 1:24714b45cd1b | 45 | */ |
xorjoep | 1:24714b45cd1b | 46 | |
xorjoep | 1:24714b45cd1b | 47 | |
xorjoep | 1:24714b45cd1b | 48 | /** |
xorjoep | 1:24714b45cd1b | 49 | * @addtogroup CmplxMatrixMult |
xorjoep | 1:24714b45cd1b | 50 | * @{ |
xorjoep | 1:24714b45cd1b | 51 | */ |
xorjoep | 1:24714b45cd1b | 52 | |
xorjoep | 1:24714b45cd1b | 53 | /** |
xorjoep | 1:24714b45cd1b | 54 | * @brief Floating-point Complex matrix multiplication. |
xorjoep | 1:24714b45cd1b | 55 | * @param[in] *pSrcA points to the first input complex matrix structure |
xorjoep | 1:24714b45cd1b | 56 | * @param[in] *pSrcB points to the second input complex matrix structure |
xorjoep | 1:24714b45cd1b | 57 | * @param[out] *pDst points to output complex matrix structure |
xorjoep | 1:24714b45cd1b | 58 | * @return The function returns either |
xorjoep | 1:24714b45cd1b | 59 | * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking. |
xorjoep | 1:24714b45cd1b | 60 | */ |
xorjoep | 1:24714b45cd1b | 61 | |
xorjoep | 1:24714b45cd1b | 62 | arm_status arm_mat_cmplx_mult_f32( |
xorjoep | 1:24714b45cd1b | 63 | const arm_matrix_instance_f32 * pSrcA, |
xorjoep | 1:24714b45cd1b | 64 | const arm_matrix_instance_f32 * pSrcB, |
xorjoep | 1:24714b45cd1b | 65 | arm_matrix_instance_f32 * pDst) |
xorjoep | 1:24714b45cd1b | 66 | { |
xorjoep | 1:24714b45cd1b | 67 | float32_t *pIn1 = pSrcA->pData; /* input data matrix pointer A */ |
xorjoep | 1:24714b45cd1b | 68 | float32_t *pIn2 = pSrcB->pData; /* input data matrix pointer B */ |
xorjoep | 1:24714b45cd1b | 69 | float32_t *pInA = pSrcA->pData; /* input data matrix pointer A */ |
xorjoep | 1:24714b45cd1b | 70 | float32_t *pOut = pDst->pData; /* output data matrix pointer */ |
xorjoep | 1:24714b45cd1b | 71 | float32_t *px; /* Temporary output data matrix pointer */ |
xorjoep | 1:24714b45cd1b | 72 | uint16_t numRowsA = pSrcA->numRows; /* number of rows of input matrix A */ |
xorjoep | 1:24714b45cd1b | 73 | uint16_t numColsB = pSrcB->numCols; /* number of columns of input matrix B */ |
xorjoep | 1:24714b45cd1b | 74 | uint16_t numColsA = pSrcA->numCols; /* number of columns of input matrix A */ |
xorjoep | 1:24714b45cd1b | 75 | float32_t sumReal1, sumImag1; /* accumulator */ |
xorjoep | 1:24714b45cd1b | 76 | float32_t a0, b0, c0, d0; |
xorjoep | 1:24714b45cd1b | 77 | float32_t a1, b1, c1, d1; |
xorjoep | 1:24714b45cd1b | 78 | float32_t sumReal2, sumImag2; /* accumulator */ |
xorjoep | 1:24714b45cd1b | 79 | |
xorjoep | 1:24714b45cd1b | 80 | |
xorjoep | 1:24714b45cd1b | 81 | /* Run the below code for Cortex-M4 and Cortex-M3 */ |
xorjoep | 1:24714b45cd1b | 82 | |
xorjoep | 1:24714b45cd1b | 83 | uint16_t col, i = 0U, j, row = numRowsA, colCnt; /* loop counters */ |
xorjoep | 1:24714b45cd1b | 84 | arm_status status; /* status of matrix multiplication */ |
xorjoep | 1:24714b45cd1b | 85 | |
xorjoep | 1:24714b45cd1b | 86 | #ifdef ARM_MATH_MATRIX_CHECK |
xorjoep | 1:24714b45cd1b | 87 | |
xorjoep | 1:24714b45cd1b | 88 | |
xorjoep | 1:24714b45cd1b | 89 | /* Check for matrix mismatch condition */ |
xorjoep | 1:24714b45cd1b | 90 | if ((pSrcA->numCols != pSrcB->numRows) || |
xorjoep | 1:24714b45cd1b | 91 | (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols)) |
xorjoep | 1:24714b45cd1b | 92 | { |
xorjoep | 1:24714b45cd1b | 93 | |
xorjoep | 1:24714b45cd1b | 94 | /* Set status as ARM_MATH_SIZE_MISMATCH */ |
xorjoep | 1:24714b45cd1b | 95 | status = ARM_MATH_SIZE_MISMATCH; |
xorjoep | 1:24714b45cd1b | 96 | } |
xorjoep | 1:24714b45cd1b | 97 | else |
xorjoep | 1:24714b45cd1b | 98 | #endif /* #ifdef ARM_MATH_MATRIX_CHECK */ |
xorjoep | 1:24714b45cd1b | 99 | |
xorjoep | 1:24714b45cd1b | 100 | { |
xorjoep | 1:24714b45cd1b | 101 | /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */ |
xorjoep | 1:24714b45cd1b | 102 | /* row loop */ |
xorjoep | 1:24714b45cd1b | 103 | do |
xorjoep | 1:24714b45cd1b | 104 | { |
xorjoep | 1:24714b45cd1b | 105 | /* Output pointer is set to starting address of the row being processed */ |
xorjoep | 1:24714b45cd1b | 106 | px = pOut + 2 * i; |
xorjoep | 1:24714b45cd1b | 107 | |
xorjoep | 1:24714b45cd1b | 108 | /* For every row wise process, the column loop counter is to be initiated */ |
xorjoep | 1:24714b45cd1b | 109 | col = numColsB; |
xorjoep | 1:24714b45cd1b | 110 | |
xorjoep | 1:24714b45cd1b | 111 | /* For every row wise process, the pIn2 pointer is set |
xorjoep | 1:24714b45cd1b | 112 | ** to the starting address of the pSrcB data */ |
xorjoep | 1:24714b45cd1b | 113 | pIn2 = pSrcB->pData; |
xorjoep | 1:24714b45cd1b | 114 | |
xorjoep | 1:24714b45cd1b | 115 | j = 0U; |
xorjoep | 1:24714b45cd1b | 116 | |
xorjoep | 1:24714b45cd1b | 117 | /* column loop */ |
xorjoep | 1:24714b45cd1b | 118 | do |
xorjoep | 1:24714b45cd1b | 119 | { |
xorjoep | 1:24714b45cd1b | 120 | /* Set the variable sum, that acts as accumulator, to zero */ |
xorjoep | 1:24714b45cd1b | 121 | sumReal1 = 0.0f; |
xorjoep | 1:24714b45cd1b | 122 | sumImag1 = 0.0f; |
xorjoep | 1:24714b45cd1b | 123 | |
xorjoep | 1:24714b45cd1b | 124 | sumReal2 = 0.0f; |
xorjoep | 1:24714b45cd1b | 125 | sumImag2 = 0.0f; |
xorjoep | 1:24714b45cd1b | 126 | |
xorjoep | 1:24714b45cd1b | 127 | /* Initiate the pointer pIn1 to point to the starting address of the column being processed */ |
xorjoep | 1:24714b45cd1b | 128 | pIn1 = pInA; |
xorjoep | 1:24714b45cd1b | 129 | |
xorjoep | 1:24714b45cd1b | 130 | /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
xorjoep | 1:24714b45cd1b | 131 | colCnt = numColsA >> 2; |
xorjoep | 1:24714b45cd1b | 132 | |
xorjoep | 1:24714b45cd1b | 133 | /* matrix multiplication */ |
xorjoep | 1:24714b45cd1b | 134 | while (colCnt > 0U) |
xorjoep | 1:24714b45cd1b | 135 | { |
xorjoep | 1:24714b45cd1b | 136 | |
xorjoep | 1:24714b45cd1b | 137 | /* Reading real part of complex matrix A */ |
xorjoep | 1:24714b45cd1b | 138 | a0 = *pIn1; |
xorjoep | 1:24714b45cd1b | 139 | |
xorjoep | 1:24714b45cd1b | 140 | /* Reading real part of complex matrix B */ |
xorjoep | 1:24714b45cd1b | 141 | c0 = *pIn2; |
xorjoep | 1:24714b45cd1b | 142 | |
xorjoep | 1:24714b45cd1b | 143 | /* Reading imaginary part of complex matrix A */ |
xorjoep | 1:24714b45cd1b | 144 | b0 = *(pIn1 + 1U); |
xorjoep | 1:24714b45cd1b | 145 | |
xorjoep | 1:24714b45cd1b | 146 | /* Reading imaginary part of complex matrix B */ |
xorjoep | 1:24714b45cd1b | 147 | d0 = *(pIn2 + 1U); |
xorjoep | 1:24714b45cd1b | 148 | |
xorjoep | 1:24714b45cd1b | 149 | sumReal1 += a0 * c0; |
xorjoep | 1:24714b45cd1b | 150 | sumImag1 += b0 * c0; |
xorjoep | 1:24714b45cd1b | 151 | |
xorjoep | 1:24714b45cd1b | 152 | pIn1 += 2U; |
xorjoep | 1:24714b45cd1b | 153 | pIn2 += 2 * numColsB; |
xorjoep | 1:24714b45cd1b | 154 | |
xorjoep | 1:24714b45cd1b | 155 | sumReal2 -= b0 * d0; |
xorjoep | 1:24714b45cd1b | 156 | sumImag2 += a0 * d0; |
xorjoep | 1:24714b45cd1b | 157 | |
xorjoep | 1:24714b45cd1b | 158 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
xorjoep | 1:24714b45cd1b | 159 | |
xorjoep | 1:24714b45cd1b | 160 | a1 = *pIn1; |
xorjoep | 1:24714b45cd1b | 161 | c1 = *pIn2; |
xorjoep | 1:24714b45cd1b | 162 | |
xorjoep | 1:24714b45cd1b | 163 | b1 = *(pIn1 + 1U); |
xorjoep | 1:24714b45cd1b | 164 | d1 = *(pIn2 + 1U); |
xorjoep | 1:24714b45cd1b | 165 | |
xorjoep | 1:24714b45cd1b | 166 | sumReal1 += a1 * c1; |
xorjoep | 1:24714b45cd1b | 167 | sumImag1 += b1 * c1; |
xorjoep | 1:24714b45cd1b | 168 | |
xorjoep | 1:24714b45cd1b | 169 | pIn1 += 2U; |
xorjoep | 1:24714b45cd1b | 170 | pIn2 += 2 * numColsB; |
xorjoep | 1:24714b45cd1b | 171 | |
xorjoep | 1:24714b45cd1b | 172 | sumReal2 -= b1 * d1; |
xorjoep | 1:24714b45cd1b | 173 | sumImag2 += a1 * d1; |
xorjoep | 1:24714b45cd1b | 174 | |
xorjoep | 1:24714b45cd1b | 175 | a0 = *pIn1; |
xorjoep | 1:24714b45cd1b | 176 | c0 = *pIn2; |
xorjoep | 1:24714b45cd1b | 177 | |
xorjoep | 1:24714b45cd1b | 178 | b0 = *(pIn1 + 1U); |
xorjoep | 1:24714b45cd1b | 179 | d0 = *(pIn2 + 1U); |
xorjoep | 1:24714b45cd1b | 180 | |
xorjoep | 1:24714b45cd1b | 181 | sumReal1 += a0 * c0; |
xorjoep | 1:24714b45cd1b | 182 | sumImag1 += b0 * c0; |
xorjoep | 1:24714b45cd1b | 183 | |
xorjoep | 1:24714b45cd1b | 184 | pIn1 += 2U; |
xorjoep | 1:24714b45cd1b | 185 | pIn2 += 2 * numColsB; |
xorjoep | 1:24714b45cd1b | 186 | |
xorjoep | 1:24714b45cd1b | 187 | sumReal2 -= b0 * d0; |
xorjoep | 1:24714b45cd1b | 188 | sumImag2 += a0 * d0; |
xorjoep | 1:24714b45cd1b | 189 | |
xorjoep | 1:24714b45cd1b | 190 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
xorjoep | 1:24714b45cd1b | 191 | |
xorjoep | 1:24714b45cd1b | 192 | a1 = *pIn1; |
xorjoep | 1:24714b45cd1b | 193 | c1 = *pIn2; |
xorjoep | 1:24714b45cd1b | 194 | |
xorjoep | 1:24714b45cd1b | 195 | b1 = *(pIn1 + 1U); |
xorjoep | 1:24714b45cd1b | 196 | d1 = *(pIn2 + 1U); |
xorjoep | 1:24714b45cd1b | 197 | |
xorjoep | 1:24714b45cd1b | 198 | sumReal1 += a1 * c1; |
xorjoep | 1:24714b45cd1b | 199 | sumImag1 += b1 * c1; |
xorjoep | 1:24714b45cd1b | 200 | |
xorjoep | 1:24714b45cd1b | 201 | pIn1 += 2U; |
xorjoep | 1:24714b45cd1b | 202 | pIn2 += 2 * numColsB; |
xorjoep | 1:24714b45cd1b | 203 | |
xorjoep | 1:24714b45cd1b | 204 | sumReal2 -= b1 * d1; |
xorjoep | 1:24714b45cd1b | 205 | sumImag2 += a1 * d1; |
xorjoep | 1:24714b45cd1b | 206 | |
xorjoep | 1:24714b45cd1b | 207 | /* Decrement the loop count */ |
xorjoep | 1:24714b45cd1b | 208 | colCnt--; |
xorjoep | 1:24714b45cd1b | 209 | } |
xorjoep | 1:24714b45cd1b | 210 | |
xorjoep | 1:24714b45cd1b | 211 | /* If the columns of pSrcA is not a multiple of 4, compute any remaining MACs here. |
xorjoep | 1:24714b45cd1b | 212 | ** No loop unrolling is used. */ |
xorjoep | 1:24714b45cd1b | 213 | colCnt = numColsA % 0x4U; |
xorjoep | 1:24714b45cd1b | 214 | |
xorjoep | 1:24714b45cd1b | 215 | while (colCnt > 0U) |
xorjoep | 1:24714b45cd1b | 216 | { |
xorjoep | 1:24714b45cd1b | 217 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
xorjoep | 1:24714b45cd1b | 218 | a1 = *pIn1; |
xorjoep | 1:24714b45cd1b | 219 | c1 = *pIn2; |
xorjoep | 1:24714b45cd1b | 220 | |
xorjoep | 1:24714b45cd1b | 221 | b1 = *(pIn1 + 1U); |
xorjoep | 1:24714b45cd1b | 222 | d1 = *(pIn2 + 1U); |
xorjoep | 1:24714b45cd1b | 223 | |
xorjoep | 1:24714b45cd1b | 224 | sumReal1 += a1 * c1; |
xorjoep | 1:24714b45cd1b | 225 | sumImag1 += b1 * c1; |
xorjoep | 1:24714b45cd1b | 226 | |
xorjoep | 1:24714b45cd1b | 227 | pIn1 += 2U; |
xorjoep | 1:24714b45cd1b | 228 | pIn2 += 2 * numColsB; |
xorjoep | 1:24714b45cd1b | 229 | |
xorjoep | 1:24714b45cd1b | 230 | sumReal2 -= b1 * d1; |
xorjoep | 1:24714b45cd1b | 231 | sumImag2 += a1 * d1; |
xorjoep | 1:24714b45cd1b | 232 | |
xorjoep | 1:24714b45cd1b | 233 | /* Decrement the loop counter */ |
xorjoep | 1:24714b45cd1b | 234 | colCnt--; |
xorjoep | 1:24714b45cd1b | 235 | } |
xorjoep | 1:24714b45cd1b | 236 | |
xorjoep | 1:24714b45cd1b | 237 | sumReal1 += sumReal2; |
xorjoep | 1:24714b45cd1b | 238 | sumImag1 += sumImag2; |
xorjoep | 1:24714b45cd1b | 239 | |
xorjoep | 1:24714b45cd1b | 240 | /* Store the result in the destination buffer */ |
xorjoep | 1:24714b45cd1b | 241 | *px++ = sumReal1; |
xorjoep | 1:24714b45cd1b | 242 | *px++ = sumImag1; |
xorjoep | 1:24714b45cd1b | 243 | |
xorjoep | 1:24714b45cd1b | 244 | /* Update the pointer pIn2 to point to the starting address of the next column */ |
xorjoep | 1:24714b45cd1b | 245 | j++; |
xorjoep | 1:24714b45cd1b | 246 | pIn2 = pSrcB->pData + 2U * j; |
xorjoep | 1:24714b45cd1b | 247 | |
xorjoep | 1:24714b45cd1b | 248 | /* Decrement the column loop counter */ |
xorjoep | 1:24714b45cd1b | 249 | col--; |
xorjoep | 1:24714b45cd1b | 250 | |
xorjoep | 1:24714b45cd1b | 251 | } while (col > 0U); |
xorjoep | 1:24714b45cd1b | 252 | |
xorjoep | 1:24714b45cd1b | 253 | /* Update the pointer pInA to point to the starting address of the next row */ |
xorjoep | 1:24714b45cd1b | 254 | i = i + numColsB; |
xorjoep | 1:24714b45cd1b | 255 | pInA = pInA + 2 * numColsA; |
xorjoep | 1:24714b45cd1b | 256 | |
xorjoep | 1:24714b45cd1b | 257 | /* Decrement the row loop counter */ |
xorjoep | 1:24714b45cd1b | 258 | row--; |
xorjoep | 1:24714b45cd1b | 259 | |
xorjoep | 1:24714b45cd1b | 260 | } while (row > 0U); |
xorjoep | 1:24714b45cd1b | 261 | |
xorjoep | 1:24714b45cd1b | 262 | /* Set status as ARM_MATH_SUCCESS */ |
xorjoep | 1:24714b45cd1b | 263 | status = ARM_MATH_SUCCESS; |
xorjoep | 1:24714b45cd1b | 264 | } |
xorjoep | 1:24714b45cd1b | 265 | |
xorjoep | 1:24714b45cd1b | 266 | /* Return to application */ |
xorjoep | 1:24714b45cd1b | 267 | return (status); |
xorjoep | 1:24714b45cd1b | 268 | } |
xorjoep | 1:24714b45cd1b | 269 | |
xorjoep | 1:24714b45cd1b | 270 | /** |
xorjoep | 1:24714b45cd1b | 271 | * @} end of MatrixMult group |
xorjoep | 1:24714b45cd1b | 272 | */ |