The CMSIS DSP 5 library
Dependents: Nucleo-Heart-Rate ejercicioVrms2 PROYECTOFINAL ejercicioVrms ... more
functions/MatrixFunctions/arm_mat_mult_fast_q31.c@3:4098b9d3d571, 2018-06-21 (annotated)
- Committer:
- xorjoep
- Date:
- Thu Jun 21 11:56:27 2018 +0000
- Revision:
- 3:4098b9d3d571
- Parent:
- 1:24714b45cd1b
headers is a folder not a 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_fast_q31.c |
xorjoep | 1:24714b45cd1b | 4 | * Description: Q31 matrix multiplication (fast variant) |
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 | * @brief Q31 matrix multiplication (fast variant) for Cortex-M3 and Cortex-M4 |
xorjoep | 1:24714b45cd1b | 42 | * @param[in] *pSrcA points to the first input matrix structure |
xorjoep | 1:24714b45cd1b | 43 | * @param[in] *pSrcB points to the second input matrix structure |
xorjoep | 1:24714b45cd1b | 44 | * @param[out] *pDst points to output matrix structure |
xorjoep | 1:24714b45cd1b | 45 | * @return The function returns either |
xorjoep | 1:24714b45cd1b | 46 | * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking. |
xorjoep | 1:24714b45cd1b | 47 | * |
xorjoep | 1:24714b45cd1b | 48 | * @details |
xorjoep | 1:24714b45cd1b | 49 | * <b>Scaling and Overflow Behavior:</b> |
xorjoep | 1:24714b45cd1b | 50 | * |
xorjoep | 1:24714b45cd1b | 51 | * \par |
xorjoep | 1:24714b45cd1b | 52 | * The difference between the function arm_mat_mult_q31() and this fast variant is that |
xorjoep | 1:24714b45cd1b | 53 | * the fast variant use a 32-bit rather than a 64-bit accumulator. |
xorjoep | 1:24714b45cd1b | 54 | * The result of each 1.31 x 1.31 multiplication is truncated to |
xorjoep | 1:24714b45cd1b | 55 | * 2.30 format. These intermediate results are accumulated in a 32-bit register in 2.30 |
xorjoep | 1:24714b45cd1b | 56 | * format. Finally, the accumulator is saturated and converted to a 1.31 result. |
xorjoep | 1:24714b45cd1b | 57 | * |
xorjoep | 1:24714b45cd1b | 58 | * \par |
xorjoep | 1:24714b45cd1b | 59 | * The fast version has the same overflow behavior as the standard version but provides |
xorjoep | 1:24714b45cd1b | 60 | * less precision since it discards the low 32 bits of each multiplication result. |
xorjoep | 1:24714b45cd1b | 61 | * In order to avoid overflows completely the input signals must be scaled down. |
xorjoep | 1:24714b45cd1b | 62 | * Scale down one of the input matrices by log2(numColsA) bits to |
xorjoep | 1:24714b45cd1b | 63 | * avoid overflows, as a total of numColsA additions are computed internally for each |
xorjoep | 1:24714b45cd1b | 64 | * output element. |
xorjoep | 1:24714b45cd1b | 65 | * |
xorjoep | 1:24714b45cd1b | 66 | * \par |
xorjoep | 1:24714b45cd1b | 67 | * See <code>arm_mat_mult_q31()</code> for a slower implementation of this function |
xorjoep | 1:24714b45cd1b | 68 | * which uses 64-bit accumulation to provide higher precision. |
xorjoep | 1:24714b45cd1b | 69 | */ |
xorjoep | 1:24714b45cd1b | 70 | |
xorjoep | 1:24714b45cd1b | 71 | arm_status arm_mat_mult_fast_q31( |
xorjoep | 1:24714b45cd1b | 72 | const arm_matrix_instance_q31 * pSrcA, |
xorjoep | 1:24714b45cd1b | 73 | const arm_matrix_instance_q31 * pSrcB, |
xorjoep | 1:24714b45cd1b | 74 | arm_matrix_instance_q31 * pDst) |
xorjoep | 1:24714b45cd1b | 75 | { |
xorjoep | 1:24714b45cd1b | 76 | q31_t *pInA = pSrcA->pData; /* input data matrix pointer A */ |
xorjoep | 1:24714b45cd1b | 77 | q31_t *pInB = pSrcB->pData; /* input data matrix pointer B */ |
xorjoep | 1:24714b45cd1b | 78 | q31_t *px; /* Temporary output data matrix pointer */ |
xorjoep | 1:24714b45cd1b | 79 | q31_t sum; /* Accumulator */ |
xorjoep | 1:24714b45cd1b | 80 | uint16_t numRowsA = pSrcA->numRows; /* number of rows of input matrix A */ |
xorjoep | 1:24714b45cd1b | 81 | uint16_t numColsB = pSrcB->numCols; /* number of columns of input matrix B */ |
xorjoep | 1:24714b45cd1b | 82 | uint16_t numColsA = pSrcA->numCols; /* number of columns of input matrix A */ |
xorjoep | 1:24714b45cd1b | 83 | uint32_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 | q31_t inA1, inB1; |
xorjoep | 1:24714b45cd1b | 86 | |
xorjoep | 1:24714b45cd1b | 87 | #if defined (ARM_MATH_DSP) |
xorjoep | 1:24714b45cd1b | 88 | |
xorjoep | 1:24714b45cd1b | 89 | q31_t sum2, sum3, sum4; |
xorjoep | 1:24714b45cd1b | 90 | q31_t inA2, inB2; |
xorjoep | 1:24714b45cd1b | 91 | q31_t *pInA2; |
xorjoep | 1:24714b45cd1b | 92 | q31_t *px2; |
xorjoep | 1:24714b45cd1b | 93 | |
xorjoep | 1:24714b45cd1b | 94 | #endif |
xorjoep | 1:24714b45cd1b | 95 | |
xorjoep | 1:24714b45cd1b | 96 | #ifdef ARM_MATH_MATRIX_CHECK |
xorjoep | 1:24714b45cd1b | 97 | |
xorjoep | 1:24714b45cd1b | 98 | /* Check for matrix mismatch condition */ |
xorjoep | 1:24714b45cd1b | 99 | if ((pSrcA->numCols != pSrcB->numRows) || |
xorjoep | 1:24714b45cd1b | 100 | (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols)) |
xorjoep | 1:24714b45cd1b | 101 | { |
xorjoep | 1:24714b45cd1b | 102 | /* Set status as ARM_MATH_SIZE_MISMATCH */ |
xorjoep | 1:24714b45cd1b | 103 | status = ARM_MATH_SIZE_MISMATCH; |
xorjoep | 1:24714b45cd1b | 104 | } |
xorjoep | 1:24714b45cd1b | 105 | else |
xorjoep | 1:24714b45cd1b | 106 | #endif /* #ifdef ARM_MATH_MATRIX_CHECK */ |
xorjoep | 1:24714b45cd1b | 107 | |
xorjoep | 1:24714b45cd1b | 108 | { |
xorjoep | 1:24714b45cd1b | 109 | |
xorjoep | 1:24714b45cd1b | 110 | px = pDst->pData; |
xorjoep | 1:24714b45cd1b | 111 | |
xorjoep | 1:24714b45cd1b | 112 | #if defined (ARM_MATH_DSP) |
xorjoep | 1:24714b45cd1b | 113 | row = row >> 1; |
xorjoep | 1:24714b45cd1b | 114 | px2 = px + numColsB; |
xorjoep | 1:24714b45cd1b | 115 | #endif |
xorjoep | 1:24714b45cd1b | 116 | |
xorjoep | 1:24714b45cd1b | 117 | /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */ |
xorjoep | 1:24714b45cd1b | 118 | /* row loop */ |
xorjoep | 1:24714b45cd1b | 119 | while (row > 0U) |
xorjoep | 1:24714b45cd1b | 120 | { |
xorjoep | 1:24714b45cd1b | 121 | |
xorjoep | 1:24714b45cd1b | 122 | /* For every row wise process, the column loop counter is to be initiated */ |
xorjoep | 1:24714b45cd1b | 123 | col = numColsB; |
xorjoep | 1:24714b45cd1b | 124 | |
xorjoep | 1:24714b45cd1b | 125 | /* For every row wise process, the pIn2 pointer is set |
xorjoep | 1:24714b45cd1b | 126 | ** to the starting address of the pSrcB data */ |
xorjoep | 1:24714b45cd1b | 127 | pInB = pSrcB->pData; |
xorjoep | 1:24714b45cd1b | 128 | |
xorjoep | 1:24714b45cd1b | 129 | j = 0U; |
xorjoep | 1:24714b45cd1b | 130 | |
xorjoep | 1:24714b45cd1b | 131 | #if defined (ARM_MATH_DSP) |
xorjoep | 1:24714b45cd1b | 132 | col = col >> 1; |
xorjoep | 1:24714b45cd1b | 133 | #endif |
xorjoep | 1:24714b45cd1b | 134 | |
xorjoep | 1:24714b45cd1b | 135 | /* column loop */ |
xorjoep | 1:24714b45cd1b | 136 | while (col > 0U) |
xorjoep | 1:24714b45cd1b | 137 | { |
xorjoep | 1:24714b45cd1b | 138 | /* Set the variable sum, that acts as accumulator, to zero */ |
xorjoep | 1:24714b45cd1b | 139 | sum = 0; |
xorjoep | 1:24714b45cd1b | 140 | |
xorjoep | 1:24714b45cd1b | 141 | /* Initiate data pointers */ |
xorjoep | 1:24714b45cd1b | 142 | pInA = pSrcA->pData + i; |
xorjoep | 1:24714b45cd1b | 143 | pInB = pSrcB->pData + j; |
xorjoep | 1:24714b45cd1b | 144 | |
xorjoep | 1:24714b45cd1b | 145 | #if defined (ARM_MATH_DSP) |
xorjoep | 1:24714b45cd1b | 146 | sum2 = 0; |
xorjoep | 1:24714b45cd1b | 147 | sum3 = 0; |
xorjoep | 1:24714b45cd1b | 148 | sum4 = 0; |
xorjoep | 1:24714b45cd1b | 149 | pInA2 = pInA + numColsA; |
xorjoep | 1:24714b45cd1b | 150 | colCnt = numColsA; |
xorjoep | 1:24714b45cd1b | 151 | #else |
xorjoep | 1:24714b45cd1b | 152 | colCnt = numColsA >> 2; |
xorjoep | 1:24714b45cd1b | 153 | #endif |
xorjoep | 1:24714b45cd1b | 154 | |
xorjoep | 1:24714b45cd1b | 155 | /* matrix multiplication */ |
xorjoep | 1:24714b45cd1b | 156 | while (colCnt > 0U) |
xorjoep | 1:24714b45cd1b | 157 | { |
xorjoep | 1:24714b45cd1b | 158 | |
xorjoep | 1:24714b45cd1b | 159 | #if defined (ARM_MATH_DSP) |
xorjoep | 1:24714b45cd1b | 160 | inA1 = *pInA++; |
xorjoep | 1:24714b45cd1b | 161 | inB1 = pInB[0]; |
xorjoep | 1:24714b45cd1b | 162 | inA2 = *pInA2++; |
xorjoep | 1:24714b45cd1b | 163 | inB2 = pInB[1]; |
xorjoep | 1:24714b45cd1b | 164 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 165 | |
xorjoep | 1:24714b45cd1b | 166 | sum = __SMMLA(inA1, inB1, sum); |
xorjoep | 1:24714b45cd1b | 167 | sum2 = __SMMLA(inA1, inB2, sum2); |
xorjoep | 1:24714b45cd1b | 168 | sum3 = __SMMLA(inA2, inB1, sum3); |
xorjoep | 1:24714b45cd1b | 169 | sum4 = __SMMLA(inA2, inB2, sum4); |
xorjoep | 1:24714b45cd1b | 170 | #else |
xorjoep | 1:24714b45cd1b | 171 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
xorjoep | 1:24714b45cd1b | 172 | /* Perform the multiply-accumulates */ |
xorjoep | 1:24714b45cd1b | 173 | inB1 = *pInB; |
xorjoep | 1:24714b45cd1b | 174 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 175 | inA1 = pInA[0]; |
xorjoep | 1:24714b45cd1b | 176 | sum = __SMMLA(inA1, inB1, sum); |
xorjoep | 1:24714b45cd1b | 177 | |
xorjoep | 1:24714b45cd1b | 178 | inB1 = *pInB; |
xorjoep | 1:24714b45cd1b | 179 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 180 | inA1 = pInA[1]; |
xorjoep | 1:24714b45cd1b | 181 | sum = __SMMLA(inA1, inB1, sum); |
xorjoep | 1:24714b45cd1b | 182 | |
xorjoep | 1:24714b45cd1b | 183 | inB1 = *pInB; |
xorjoep | 1:24714b45cd1b | 184 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 185 | inA1 = pInA[2]; |
xorjoep | 1:24714b45cd1b | 186 | sum = __SMMLA(inA1, inB1, sum); |
xorjoep | 1:24714b45cd1b | 187 | |
xorjoep | 1:24714b45cd1b | 188 | inB1 = *pInB; |
xorjoep | 1:24714b45cd1b | 189 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 190 | inA1 = pInA[3]; |
xorjoep | 1:24714b45cd1b | 191 | sum = __SMMLA(inA1, inB1, sum); |
xorjoep | 1:24714b45cd1b | 192 | |
xorjoep | 1:24714b45cd1b | 193 | pInA += 4U; |
xorjoep | 1:24714b45cd1b | 194 | #endif |
xorjoep | 1:24714b45cd1b | 195 | |
xorjoep | 1:24714b45cd1b | 196 | /* Decrement the loop counter */ |
xorjoep | 1:24714b45cd1b | 197 | colCnt--; |
xorjoep | 1:24714b45cd1b | 198 | } |
xorjoep | 1:24714b45cd1b | 199 | |
xorjoep | 1:24714b45cd1b | 200 | #ifdef ARM_MATH_CM0_FAMILY |
xorjoep | 1:24714b45cd1b | 201 | /* If the columns of pSrcA is not a multiple of 4, compute any remaining output samples here. */ |
xorjoep | 1:24714b45cd1b | 202 | colCnt = numColsA % 0x4U; |
xorjoep | 1:24714b45cd1b | 203 | while (colCnt > 0U) |
xorjoep | 1:24714b45cd1b | 204 | { |
xorjoep | 1:24714b45cd1b | 205 | sum = __SMMLA(*pInA++, *pInB, sum); |
xorjoep | 1:24714b45cd1b | 206 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 207 | colCnt--; |
xorjoep | 1:24714b45cd1b | 208 | } |
xorjoep | 1:24714b45cd1b | 209 | j++; |
xorjoep | 1:24714b45cd1b | 210 | #endif |
xorjoep | 1:24714b45cd1b | 211 | |
xorjoep | 1:24714b45cd1b | 212 | /* Convert the result from 2.30 to 1.31 format and store in destination buffer */ |
xorjoep | 1:24714b45cd1b | 213 | *px++ = sum << 1; |
xorjoep | 1:24714b45cd1b | 214 | |
xorjoep | 1:24714b45cd1b | 215 | #if defined (ARM_MATH_DSP) |
xorjoep | 1:24714b45cd1b | 216 | *px++ = sum2 << 1; |
xorjoep | 1:24714b45cd1b | 217 | *px2++ = sum3 << 1; |
xorjoep | 1:24714b45cd1b | 218 | *px2++ = sum4 << 1; |
xorjoep | 1:24714b45cd1b | 219 | j += 2; |
xorjoep | 1:24714b45cd1b | 220 | #endif |
xorjoep | 1:24714b45cd1b | 221 | |
xorjoep | 1:24714b45cd1b | 222 | /* Decrement the column loop counter */ |
xorjoep | 1:24714b45cd1b | 223 | col--; |
xorjoep | 1:24714b45cd1b | 224 | |
xorjoep | 1:24714b45cd1b | 225 | } |
xorjoep | 1:24714b45cd1b | 226 | |
xorjoep | 1:24714b45cd1b | 227 | i = i + numColsA; |
xorjoep | 1:24714b45cd1b | 228 | |
xorjoep | 1:24714b45cd1b | 229 | #if defined (ARM_MATH_DSP) |
xorjoep | 1:24714b45cd1b | 230 | i = i + numColsA; |
xorjoep | 1:24714b45cd1b | 231 | px = px2 + (numColsB & 1U); |
xorjoep | 1:24714b45cd1b | 232 | px2 = px + numColsB; |
xorjoep | 1:24714b45cd1b | 233 | #endif |
xorjoep | 1:24714b45cd1b | 234 | |
xorjoep | 1:24714b45cd1b | 235 | /* Decrement the row loop counter */ |
xorjoep | 1:24714b45cd1b | 236 | row--; |
xorjoep | 1:24714b45cd1b | 237 | |
xorjoep | 1:24714b45cd1b | 238 | } |
xorjoep | 1:24714b45cd1b | 239 | |
xorjoep | 1:24714b45cd1b | 240 | /* Compute any remaining odd row/column below */ |
xorjoep | 1:24714b45cd1b | 241 | |
xorjoep | 1:24714b45cd1b | 242 | #if defined (ARM_MATH_DSP) |
xorjoep | 1:24714b45cd1b | 243 | |
xorjoep | 1:24714b45cd1b | 244 | /* Compute remaining output column */ |
xorjoep | 1:24714b45cd1b | 245 | if (numColsB & 1U) { |
xorjoep | 1:24714b45cd1b | 246 | |
xorjoep | 1:24714b45cd1b | 247 | /* Avoid redundant computation of last element */ |
xorjoep | 1:24714b45cd1b | 248 | row = numRowsA & (~0x1); |
xorjoep | 1:24714b45cd1b | 249 | |
xorjoep | 1:24714b45cd1b | 250 | /* Point to remaining unfilled column in output matrix */ |
xorjoep | 1:24714b45cd1b | 251 | px = pDst->pData+numColsB-1; |
xorjoep | 1:24714b45cd1b | 252 | pInA = pSrcA->pData; |
xorjoep | 1:24714b45cd1b | 253 | |
xorjoep | 1:24714b45cd1b | 254 | /* row loop */ |
xorjoep | 1:24714b45cd1b | 255 | while (row > 0) |
xorjoep | 1:24714b45cd1b | 256 | { |
xorjoep | 1:24714b45cd1b | 257 | |
xorjoep | 1:24714b45cd1b | 258 | /* point to last column in matrix B */ |
xorjoep | 1:24714b45cd1b | 259 | pInB = pSrcB->pData + numColsB-1; |
xorjoep | 1:24714b45cd1b | 260 | |
xorjoep | 1:24714b45cd1b | 261 | /* Set the variable sum, that acts as accumulator, to zero */ |
xorjoep | 1:24714b45cd1b | 262 | sum = 0; |
xorjoep | 1:24714b45cd1b | 263 | |
xorjoep | 1:24714b45cd1b | 264 | /* Compute 4 columns at once */ |
xorjoep | 1:24714b45cd1b | 265 | colCnt = numColsA >> 2; |
xorjoep | 1:24714b45cd1b | 266 | |
xorjoep | 1:24714b45cd1b | 267 | /* matrix multiplication */ |
xorjoep | 1:24714b45cd1b | 268 | while (colCnt > 0U) |
xorjoep | 1:24714b45cd1b | 269 | { |
xorjoep | 1:24714b45cd1b | 270 | inA1 = *pInA++; |
xorjoep | 1:24714b45cd1b | 271 | inA2 = *pInA++; |
xorjoep | 1:24714b45cd1b | 272 | inB1 = *pInB; |
xorjoep | 1:24714b45cd1b | 273 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 274 | inB2 = *pInB; |
xorjoep | 1:24714b45cd1b | 275 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 276 | sum = __SMMLA(inA1, inB1, sum); |
xorjoep | 1:24714b45cd1b | 277 | sum = __SMMLA(inA2, inB2, sum); |
xorjoep | 1:24714b45cd1b | 278 | |
xorjoep | 1:24714b45cd1b | 279 | inA1 = *pInA++; |
xorjoep | 1:24714b45cd1b | 280 | inA2 = *pInA++; |
xorjoep | 1:24714b45cd1b | 281 | inB1 = *pInB; |
xorjoep | 1:24714b45cd1b | 282 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 283 | inB2 = *pInB; |
xorjoep | 1:24714b45cd1b | 284 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 285 | sum = __SMMLA(inA1, inB1, sum); |
xorjoep | 1:24714b45cd1b | 286 | sum = __SMMLA(inA2, inB2, sum); |
xorjoep | 1:24714b45cd1b | 287 | |
xorjoep | 1:24714b45cd1b | 288 | /* Decrement the loop counter */ |
xorjoep | 1:24714b45cd1b | 289 | colCnt--; |
xorjoep | 1:24714b45cd1b | 290 | } |
xorjoep | 1:24714b45cd1b | 291 | |
xorjoep | 1:24714b45cd1b | 292 | colCnt = numColsA & 3U; |
xorjoep | 1:24714b45cd1b | 293 | while (colCnt > 0U) { |
xorjoep | 1:24714b45cd1b | 294 | sum = __SMMLA(*pInA++, *pInB, sum); |
xorjoep | 1:24714b45cd1b | 295 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 296 | colCnt--; |
xorjoep | 1:24714b45cd1b | 297 | } |
xorjoep | 1:24714b45cd1b | 298 | |
xorjoep | 1:24714b45cd1b | 299 | /* Convert the result from 2.30 to 1.31 format and store in destination buffer */ |
xorjoep | 1:24714b45cd1b | 300 | *px = sum << 1; |
xorjoep | 1:24714b45cd1b | 301 | px += numColsB; |
xorjoep | 1:24714b45cd1b | 302 | |
xorjoep | 1:24714b45cd1b | 303 | /* Decrement the row loop counter */ |
xorjoep | 1:24714b45cd1b | 304 | row--; |
xorjoep | 1:24714b45cd1b | 305 | } |
xorjoep | 1:24714b45cd1b | 306 | } |
xorjoep | 1:24714b45cd1b | 307 | |
xorjoep | 1:24714b45cd1b | 308 | /* Compute remaining output row */ |
xorjoep | 1:24714b45cd1b | 309 | if (numRowsA & 1U) { |
xorjoep | 1:24714b45cd1b | 310 | |
xorjoep | 1:24714b45cd1b | 311 | /* point to last row in output matrix */ |
xorjoep | 1:24714b45cd1b | 312 | px = pDst->pData+(numColsB)*(numRowsA-1); |
xorjoep | 1:24714b45cd1b | 313 | |
xorjoep | 1:24714b45cd1b | 314 | col = numColsB; |
xorjoep | 1:24714b45cd1b | 315 | i = 0U; |
xorjoep | 1:24714b45cd1b | 316 | |
xorjoep | 1:24714b45cd1b | 317 | /* col loop */ |
xorjoep | 1:24714b45cd1b | 318 | while (col > 0) |
xorjoep | 1:24714b45cd1b | 319 | { |
xorjoep | 1:24714b45cd1b | 320 | |
xorjoep | 1:24714b45cd1b | 321 | /* point to last row in matrix A */ |
xorjoep | 1:24714b45cd1b | 322 | pInA = pSrcA->pData + (numRowsA-1)*numColsA; |
xorjoep | 1:24714b45cd1b | 323 | pInB = pSrcB->pData + i; |
xorjoep | 1:24714b45cd1b | 324 | |
xorjoep | 1:24714b45cd1b | 325 | /* Set the variable sum, that acts as accumulator, to zero */ |
xorjoep | 1:24714b45cd1b | 326 | sum = 0; |
xorjoep | 1:24714b45cd1b | 327 | |
xorjoep | 1:24714b45cd1b | 328 | /* Compute 4 columns at once */ |
xorjoep | 1:24714b45cd1b | 329 | colCnt = numColsA >> 2; |
xorjoep | 1:24714b45cd1b | 330 | |
xorjoep | 1:24714b45cd1b | 331 | /* matrix multiplication */ |
xorjoep | 1:24714b45cd1b | 332 | while (colCnt > 0U) |
xorjoep | 1:24714b45cd1b | 333 | { |
xorjoep | 1:24714b45cd1b | 334 | inA1 = *pInA++; |
xorjoep | 1:24714b45cd1b | 335 | inA2 = *pInA++; |
xorjoep | 1:24714b45cd1b | 336 | inB1 = *pInB; |
xorjoep | 1:24714b45cd1b | 337 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 338 | inB2 = *pInB; |
xorjoep | 1:24714b45cd1b | 339 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 340 | sum = __SMMLA(inA1, inB1, sum); |
xorjoep | 1:24714b45cd1b | 341 | sum = __SMMLA(inA2, inB2, sum); |
xorjoep | 1:24714b45cd1b | 342 | |
xorjoep | 1:24714b45cd1b | 343 | inA1 = *pInA++; |
xorjoep | 1:24714b45cd1b | 344 | inA2 = *pInA++; |
xorjoep | 1:24714b45cd1b | 345 | inB1 = *pInB; |
xorjoep | 1:24714b45cd1b | 346 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 347 | inB2 = *pInB; |
xorjoep | 1:24714b45cd1b | 348 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 349 | sum = __SMMLA(inA1, inB1, sum); |
xorjoep | 1:24714b45cd1b | 350 | sum = __SMMLA(inA2, inB2, sum); |
xorjoep | 1:24714b45cd1b | 351 | |
xorjoep | 1:24714b45cd1b | 352 | /* Decrement the loop counter */ |
xorjoep | 1:24714b45cd1b | 353 | colCnt--; |
xorjoep | 1:24714b45cd1b | 354 | } |
xorjoep | 1:24714b45cd1b | 355 | |
xorjoep | 1:24714b45cd1b | 356 | colCnt = numColsA & 3U; |
xorjoep | 1:24714b45cd1b | 357 | while (colCnt > 0U) { |
xorjoep | 1:24714b45cd1b | 358 | sum = __SMMLA(*pInA++, *pInB, sum); |
xorjoep | 1:24714b45cd1b | 359 | pInB += numColsB; |
xorjoep | 1:24714b45cd1b | 360 | colCnt--; |
xorjoep | 1:24714b45cd1b | 361 | } |
xorjoep | 1:24714b45cd1b | 362 | |
xorjoep | 1:24714b45cd1b | 363 | /* Saturate and store the result in the destination buffer */ |
xorjoep | 1:24714b45cd1b | 364 | *px++ = sum << 1; |
xorjoep | 1:24714b45cd1b | 365 | i++; |
xorjoep | 1:24714b45cd1b | 366 | |
xorjoep | 1:24714b45cd1b | 367 | /* Decrement the col loop counter */ |
xorjoep | 1:24714b45cd1b | 368 | col--; |
xorjoep | 1:24714b45cd1b | 369 | } |
xorjoep | 1:24714b45cd1b | 370 | } |
xorjoep | 1:24714b45cd1b | 371 | |
xorjoep | 1:24714b45cd1b | 372 | #endif /* #if defined (ARM_MATH_DSP) */ |
xorjoep | 1:24714b45cd1b | 373 | |
xorjoep | 1:24714b45cd1b | 374 | /* set status as ARM_MATH_SUCCESS */ |
xorjoep | 1:24714b45cd1b | 375 | status = ARM_MATH_SUCCESS; |
xorjoep | 1:24714b45cd1b | 376 | } |
xorjoep | 1:24714b45cd1b | 377 | |
xorjoep | 1:24714b45cd1b | 378 | /* Return to application */ |
xorjoep | 1:24714b45cd1b | 379 | return (status); |
xorjoep | 1:24714b45cd1b | 380 | } |
xorjoep | 1:24714b45cd1b | 381 | |
xorjoep | 1:24714b45cd1b | 382 | /** |
xorjoep | 1:24714b45cd1b | 383 | * @} end of MatrixMult group |
xorjoep | 1:24714b45cd1b | 384 | */ |