V4.0.1 of the ARM CMSIS DSP libraries. Note that arm_bitreversal2.s, arm_cfft_f32.c and arm_rfft_fast_f32.c had to be removed. arm_bitreversal2.s will not assemble with the online tools. So, the fast f32 FFT functions are not yet available. All the other FFT functions are available.
Dependents: MPU9150_Example fir_f32 fir_f32 MPU9150_nucleo_noni2cdev ... more
MatrixFunctions/arm_mat_mult_fast_q15.c@0:3d9c67d97d6f, 2014-07-28 (annotated)
- Committer:
- emh203
- Date:
- Mon Jul 28 15:03:15 2014 +0000
- Revision:
- 0:3d9c67d97d6f
1st working commit. Had to remove arm_bitreversal2.s arm_cfft_f32.c and arm_rfft_fast_f32.c. The .s will not assemble. For now I removed these functions so we could at least have a library for the other functions.
Who changed what in which revision?
User | Revision | Line number | New contents of line |
---|---|---|---|
emh203 | 0:3d9c67d97d6f | 1 | /* ---------------------------------------------------------------------- |
emh203 | 0:3d9c67d97d6f | 2 | * Copyright (C) 2010-2014 ARM Limited. All rights reserved. |
emh203 | 0:3d9c67d97d6f | 3 | * |
emh203 | 0:3d9c67d97d6f | 4 | * $Date: 12. March 2014 |
emh203 | 0:3d9c67d97d6f | 5 | * $Revision: V1.4.3 |
emh203 | 0:3d9c67d97d6f | 6 | * |
emh203 | 0:3d9c67d97d6f | 7 | * Project: CMSIS DSP Library |
emh203 | 0:3d9c67d97d6f | 8 | * Title: arm_mat_mult_fast_q15.c |
emh203 | 0:3d9c67d97d6f | 9 | * |
emh203 | 0:3d9c67d97d6f | 10 | * Description: Q15 matrix multiplication (fast variant) |
emh203 | 0:3d9c67d97d6f | 11 | * |
emh203 | 0:3d9c67d97d6f | 12 | * Target Processor: Cortex-M4/Cortex-M3 |
emh203 | 0:3d9c67d97d6f | 13 | * |
emh203 | 0:3d9c67d97d6f | 14 | * Redistribution and use in source and binary forms, with or without |
emh203 | 0:3d9c67d97d6f | 15 | * modification, are permitted provided that the following conditions |
emh203 | 0:3d9c67d97d6f | 16 | * are met: |
emh203 | 0:3d9c67d97d6f | 17 | * - Redistributions of source code must retain the above copyright |
emh203 | 0:3d9c67d97d6f | 18 | * notice, this list of conditions and the following disclaimer. |
emh203 | 0:3d9c67d97d6f | 19 | * - Redistributions in binary form must reproduce the above copyright |
emh203 | 0:3d9c67d97d6f | 20 | * notice, this list of conditions and the following disclaimer in |
emh203 | 0:3d9c67d97d6f | 21 | * the documentation and/or other materials provided with the |
emh203 | 0:3d9c67d97d6f | 22 | * distribution. |
emh203 | 0:3d9c67d97d6f | 23 | * - Neither the name of ARM LIMITED nor the names of its contributors |
emh203 | 0:3d9c67d97d6f | 24 | * may be used to endorse or promote products derived from this |
emh203 | 0:3d9c67d97d6f | 25 | * software without specific prior written permission. |
emh203 | 0:3d9c67d97d6f | 26 | * |
emh203 | 0:3d9c67d97d6f | 27 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
emh203 | 0:3d9c67d97d6f | 28 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
emh203 | 0:3d9c67d97d6f | 29 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
emh203 | 0:3d9c67d97d6f | 30 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
emh203 | 0:3d9c67d97d6f | 31 | * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
emh203 | 0:3d9c67d97d6f | 32 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
emh203 | 0:3d9c67d97d6f | 33 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
emh203 | 0:3d9c67d97d6f | 34 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
emh203 | 0:3d9c67d97d6f | 35 | * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
emh203 | 0:3d9c67d97d6f | 36 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN |
emh203 | 0:3d9c67d97d6f | 37 | * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
emh203 | 0:3d9c67d97d6f | 38 | * POSSIBILITY OF SUCH DAMAGE. |
emh203 | 0:3d9c67d97d6f | 39 | * -------------------------------------------------------------------- */ |
emh203 | 0:3d9c67d97d6f | 40 | |
emh203 | 0:3d9c67d97d6f | 41 | #include "arm_math.h" |
emh203 | 0:3d9c67d97d6f | 42 | |
emh203 | 0:3d9c67d97d6f | 43 | /** |
emh203 | 0:3d9c67d97d6f | 44 | * @ingroup groupMatrix |
emh203 | 0:3d9c67d97d6f | 45 | */ |
emh203 | 0:3d9c67d97d6f | 46 | |
emh203 | 0:3d9c67d97d6f | 47 | /** |
emh203 | 0:3d9c67d97d6f | 48 | * @addtogroup MatrixMult |
emh203 | 0:3d9c67d97d6f | 49 | * @{ |
emh203 | 0:3d9c67d97d6f | 50 | */ |
emh203 | 0:3d9c67d97d6f | 51 | |
emh203 | 0:3d9c67d97d6f | 52 | |
emh203 | 0:3d9c67d97d6f | 53 | /** |
emh203 | 0:3d9c67d97d6f | 54 | * @brief Q15 matrix multiplication (fast variant) for Cortex-M3 and Cortex-M4 |
emh203 | 0:3d9c67d97d6f | 55 | * @param[in] *pSrcA points to the first input matrix structure |
emh203 | 0:3d9c67d97d6f | 56 | * @param[in] *pSrcB points to the second input matrix structure |
emh203 | 0:3d9c67d97d6f | 57 | * @param[out] *pDst points to output matrix structure |
emh203 | 0:3d9c67d97d6f | 58 | * @param[in] *pState points to the array for storing intermediate results |
emh203 | 0:3d9c67d97d6f | 59 | * @return The function returns either |
emh203 | 0:3d9c67d97d6f | 60 | * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking. |
emh203 | 0:3d9c67d97d6f | 61 | * |
emh203 | 0:3d9c67d97d6f | 62 | * @details |
emh203 | 0:3d9c67d97d6f | 63 | * <b>Scaling and Overflow Behavior:</b> |
emh203 | 0:3d9c67d97d6f | 64 | * |
emh203 | 0:3d9c67d97d6f | 65 | * \par |
emh203 | 0:3d9c67d97d6f | 66 | * The difference between the function arm_mat_mult_q15() and this fast variant is that |
emh203 | 0:3d9c67d97d6f | 67 | * the fast variant use a 32-bit rather than a 64-bit accumulator. |
emh203 | 0:3d9c67d97d6f | 68 | * The result of each 1.15 x 1.15 multiplication is truncated to |
emh203 | 0:3d9c67d97d6f | 69 | * 2.30 format. These intermediate results are accumulated in a 32-bit register in 2.30 |
emh203 | 0:3d9c67d97d6f | 70 | * format. Finally, the accumulator is saturated and converted to a 1.15 result. |
emh203 | 0:3d9c67d97d6f | 71 | * |
emh203 | 0:3d9c67d97d6f | 72 | * \par |
emh203 | 0:3d9c67d97d6f | 73 | * The fast version has the same overflow behavior as the standard version but provides |
emh203 | 0:3d9c67d97d6f | 74 | * less precision since it discards the low 16 bits of each multiplication result. |
emh203 | 0:3d9c67d97d6f | 75 | * In order to avoid overflows completely the input signals must be scaled down. |
emh203 | 0:3d9c67d97d6f | 76 | * Scale down one of the input matrices by log2(numColsA) bits to |
emh203 | 0:3d9c67d97d6f | 77 | * avoid overflows, as a total of numColsA additions are computed internally for each |
emh203 | 0:3d9c67d97d6f | 78 | * output element. |
emh203 | 0:3d9c67d97d6f | 79 | * |
emh203 | 0:3d9c67d97d6f | 80 | * \par |
emh203 | 0:3d9c67d97d6f | 81 | * See <code>arm_mat_mult_q15()</code> for a slower implementation of this function |
emh203 | 0:3d9c67d97d6f | 82 | * which uses 64-bit accumulation to provide higher precision. |
emh203 | 0:3d9c67d97d6f | 83 | */ |
emh203 | 0:3d9c67d97d6f | 84 | |
emh203 | 0:3d9c67d97d6f | 85 | arm_status arm_mat_mult_fast_q15( |
emh203 | 0:3d9c67d97d6f | 86 | const arm_matrix_instance_q15 * pSrcA, |
emh203 | 0:3d9c67d97d6f | 87 | const arm_matrix_instance_q15 * pSrcB, |
emh203 | 0:3d9c67d97d6f | 88 | arm_matrix_instance_q15 * pDst, |
emh203 | 0:3d9c67d97d6f | 89 | q15_t * pState) |
emh203 | 0:3d9c67d97d6f | 90 | { |
emh203 | 0:3d9c67d97d6f | 91 | q31_t sum; /* accumulator */ |
emh203 | 0:3d9c67d97d6f | 92 | q15_t *pSrcBT = pState; /* input data matrix pointer for transpose */ |
emh203 | 0:3d9c67d97d6f | 93 | q15_t *pInA = pSrcA->pData; /* input data matrix pointer A of Q15 type */ |
emh203 | 0:3d9c67d97d6f | 94 | q15_t *pInB = pSrcB->pData; /* input data matrix pointer B of Q15 type */ |
emh203 | 0:3d9c67d97d6f | 95 | q15_t *px; /* Temporary output data matrix pointer */ |
emh203 | 0:3d9c67d97d6f | 96 | uint16_t numRowsA = pSrcA->numRows; /* number of rows of input matrix A */ |
emh203 | 0:3d9c67d97d6f | 97 | uint16_t numColsB = pSrcB->numCols; /* number of columns of input matrix B */ |
emh203 | 0:3d9c67d97d6f | 98 | uint16_t numColsA = pSrcA->numCols; /* number of columns of input matrix A */ |
emh203 | 0:3d9c67d97d6f | 99 | uint16_t numRowsB = pSrcB->numRows; /* number of rows of input matrix A */ |
emh203 | 0:3d9c67d97d6f | 100 | uint16_t col, i = 0u, row = numRowsB, colCnt; /* loop counters */ |
emh203 | 0:3d9c67d97d6f | 101 | arm_status status; /* status of matrix multiplication */ |
emh203 | 0:3d9c67d97d6f | 102 | |
emh203 | 0:3d9c67d97d6f | 103 | #ifndef UNALIGNED_SUPPORT_DISABLE |
emh203 | 0:3d9c67d97d6f | 104 | |
emh203 | 0:3d9c67d97d6f | 105 | q31_t in; /* Temporary variable to hold the input value */ |
emh203 | 0:3d9c67d97d6f | 106 | q31_t inA1, inA2, inB1, inB2; |
emh203 | 0:3d9c67d97d6f | 107 | |
emh203 | 0:3d9c67d97d6f | 108 | #else |
emh203 | 0:3d9c67d97d6f | 109 | |
emh203 | 0:3d9c67d97d6f | 110 | q15_t in; /* Temporary variable to hold the input value */ |
emh203 | 0:3d9c67d97d6f | 111 | q15_t inA1, inA2, inB1, inB2; |
emh203 | 0:3d9c67d97d6f | 112 | |
emh203 | 0:3d9c67d97d6f | 113 | #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */ |
emh203 | 0:3d9c67d97d6f | 114 | |
emh203 | 0:3d9c67d97d6f | 115 | #ifdef ARM_MATH_MATRIX_CHECK |
emh203 | 0:3d9c67d97d6f | 116 | /* Check for matrix mismatch condition */ |
emh203 | 0:3d9c67d97d6f | 117 | if((pSrcA->numCols != pSrcB->numRows) || |
emh203 | 0:3d9c67d97d6f | 118 | (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols)) |
emh203 | 0:3d9c67d97d6f | 119 | { |
emh203 | 0:3d9c67d97d6f | 120 | /* Set status as ARM_MATH_SIZE_MISMATCH */ |
emh203 | 0:3d9c67d97d6f | 121 | status = ARM_MATH_SIZE_MISMATCH; |
emh203 | 0:3d9c67d97d6f | 122 | } |
emh203 | 0:3d9c67d97d6f | 123 | else |
emh203 | 0:3d9c67d97d6f | 124 | #endif |
emh203 | 0:3d9c67d97d6f | 125 | { |
emh203 | 0:3d9c67d97d6f | 126 | /* Matrix transpose */ |
emh203 | 0:3d9c67d97d6f | 127 | do |
emh203 | 0:3d9c67d97d6f | 128 | { |
emh203 | 0:3d9c67d97d6f | 129 | /* Apply loop unrolling and exchange the columns with row elements */ |
emh203 | 0:3d9c67d97d6f | 130 | col = numColsB >> 2; |
emh203 | 0:3d9c67d97d6f | 131 | |
emh203 | 0:3d9c67d97d6f | 132 | /* The pointer px is set to starting address of the column being processed */ |
emh203 | 0:3d9c67d97d6f | 133 | px = pSrcBT + i; |
emh203 | 0:3d9c67d97d6f | 134 | |
emh203 | 0:3d9c67d97d6f | 135 | /* First part of the processing with loop unrolling. Compute 4 outputs at a time. |
emh203 | 0:3d9c67d97d6f | 136 | ** a second loop below computes the remaining 1 to 3 samples. */ |
emh203 | 0:3d9c67d97d6f | 137 | while(col > 0u) |
emh203 | 0:3d9c67d97d6f | 138 | { |
emh203 | 0:3d9c67d97d6f | 139 | #ifndef UNALIGNED_SUPPORT_DISABLE |
emh203 | 0:3d9c67d97d6f | 140 | /* Read two elements from the row */ |
emh203 | 0:3d9c67d97d6f | 141 | in = *__SIMD32(pInB)++; |
emh203 | 0:3d9c67d97d6f | 142 | |
emh203 | 0:3d9c67d97d6f | 143 | /* Unpack and store one element in the destination */ |
emh203 | 0:3d9c67d97d6f | 144 | #ifndef ARM_MATH_BIG_ENDIAN |
emh203 | 0:3d9c67d97d6f | 145 | |
emh203 | 0:3d9c67d97d6f | 146 | *px = (q15_t) in; |
emh203 | 0:3d9c67d97d6f | 147 | |
emh203 | 0:3d9c67d97d6f | 148 | #else |
emh203 | 0:3d9c67d97d6f | 149 | |
emh203 | 0:3d9c67d97d6f | 150 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
emh203 | 0:3d9c67d97d6f | 151 | |
emh203 | 0:3d9c67d97d6f | 152 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
emh203 | 0:3d9c67d97d6f | 153 | |
emh203 | 0:3d9c67d97d6f | 154 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emh203 | 0:3d9c67d97d6f | 155 | px += numRowsB; |
emh203 | 0:3d9c67d97d6f | 156 | |
emh203 | 0:3d9c67d97d6f | 157 | /* Unpack and store the second element in the destination */ |
emh203 | 0:3d9c67d97d6f | 158 | #ifndef ARM_MATH_BIG_ENDIAN |
emh203 | 0:3d9c67d97d6f | 159 | |
emh203 | 0:3d9c67d97d6f | 160 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
emh203 | 0:3d9c67d97d6f | 161 | |
emh203 | 0:3d9c67d97d6f | 162 | #else |
emh203 | 0:3d9c67d97d6f | 163 | |
emh203 | 0:3d9c67d97d6f | 164 | *px = (q15_t) in; |
emh203 | 0:3d9c67d97d6f | 165 | |
emh203 | 0:3d9c67d97d6f | 166 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
emh203 | 0:3d9c67d97d6f | 167 | |
emh203 | 0:3d9c67d97d6f | 168 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emh203 | 0:3d9c67d97d6f | 169 | px += numRowsB; |
emh203 | 0:3d9c67d97d6f | 170 | |
emh203 | 0:3d9c67d97d6f | 171 | /* Read two elements from the row */ |
emh203 | 0:3d9c67d97d6f | 172 | in = *__SIMD32(pInB)++; |
emh203 | 0:3d9c67d97d6f | 173 | |
emh203 | 0:3d9c67d97d6f | 174 | /* Unpack and store one element in the destination */ |
emh203 | 0:3d9c67d97d6f | 175 | #ifndef ARM_MATH_BIG_ENDIAN |
emh203 | 0:3d9c67d97d6f | 176 | |
emh203 | 0:3d9c67d97d6f | 177 | *px = (q15_t) in; |
emh203 | 0:3d9c67d97d6f | 178 | |
emh203 | 0:3d9c67d97d6f | 179 | #else |
emh203 | 0:3d9c67d97d6f | 180 | |
emh203 | 0:3d9c67d97d6f | 181 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
emh203 | 0:3d9c67d97d6f | 182 | |
emh203 | 0:3d9c67d97d6f | 183 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
emh203 | 0:3d9c67d97d6f | 184 | |
emh203 | 0:3d9c67d97d6f | 185 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emh203 | 0:3d9c67d97d6f | 186 | px += numRowsB; |
emh203 | 0:3d9c67d97d6f | 187 | |
emh203 | 0:3d9c67d97d6f | 188 | /* Unpack and store the second element in the destination */ |
emh203 | 0:3d9c67d97d6f | 189 | |
emh203 | 0:3d9c67d97d6f | 190 | #ifndef ARM_MATH_BIG_ENDIAN |
emh203 | 0:3d9c67d97d6f | 191 | |
emh203 | 0:3d9c67d97d6f | 192 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
emh203 | 0:3d9c67d97d6f | 193 | |
emh203 | 0:3d9c67d97d6f | 194 | #else |
emh203 | 0:3d9c67d97d6f | 195 | |
emh203 | 0:3d9c67d97d6f | 196 | *px = (q15_t) in; |
emh203 | 0:3d9c67d97d6f | 197 | |
emh203 | 0:3d9c67d97d6f | 198 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
emh203 | 0:3d9c67d97d6f | 199 | |
emh203 | 0:3d9c67d97d6f | 200 | #else |
emh203 | 0:3d9c67d97d6f | 201 | |
emh203 | 0:3d9c67d97d6f | 202 | /* Read one element from the row */ |
emh203 | 0:3d9c67d97d6f | 203 | in = *pInB++; |
emh203 | 0:3d9c67d97d6f | 204 | |
emh203 | 0:3d9c67d97d6f | 205 | /* Store one element in the destination */ |
emh203 | 0:3d9c67d97d6f | 206 | *px = in; |
emh203 | 0:3d9c67d97d6f | 207 | |
emh203 | 0:3d9c67d97d6f | 208 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emh203 | 0:3d9c67d97d6f | 209 | px += numRowsB; |
emh203 | 0:3d9c67d97d6f | 210 | |
emh203 | 0:3d9c67d97d6f | 211 | /* Read one element from the row */ |
emh203 | 0:3d9c67d97d6f | 212 | in = *pInB++; |
emh203 | 0:3d9c67d97d6f | 213 | |
emh203 | 0:3d9c67d97d6f | 214 | /* Store one element in the destination */ |
emh203 | 0:3d9c67d97d6f | 215 | *px = in; |
emh203 | 0:3d9c67d97d6f | 216 | |
emh203 | 0:3d9c67d97d6f | 217 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emh203 | 0:3d9c67d97d6f | 218 | px += numRowsB; |
emh203 | 0:3d9c67d97d6f | 219 | |
emh203 | 0:3d9c67d97d6f | 220 | /* Read one element from the row */ |
emh203 | 0:3d9c67d97d6f | 221 | in = *pInB++; |
emh203 | 0:3d9c67d97d6f | 222 | |
emh203 | 0:3d9c67d97d6f | 223 | /* Store one element in the destination */ |
emh203 | 0:3d9c67d97d6f | 224 | *px = in; |
emh203 | 0:3d9c67d97d6f | 225 | |
emh203 | 0:3d9c67d97d6f | 226 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emh203 | 0:3d9c67d97d6f | 227 | px += numRowsB; |
emh203 | 0:3d9c67d97d6f | 228 | |
emh203 | 0:3d9c67d97d6f | 229 | /* Read one element from the row */ |
emh203 | 0:3d9c67d97d6f | 230 | in = *pInB++; |
emh203 | 0:3d9c67d97d6f | 231 | |
emh203 | 0:3d9c67d97d6f | 232 | /* Store one element in the destination */ |
emh203 | 0:3d9c67d97d6f | 233 | *px = in; |
emh203 | 0:3d9c67d97d6f | 234 | |
emh203 | 0:3d9c67d97d6f | 235 | #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */ |
emh203 | 0:3d9c67d97d6f | 236 | |
emh203 | 0:3d9c67d97d6f | 237 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emh203 | 0:3d9c67d97d6f | 238 | px += numRowsB; |
emh203 | 0:3d9c67d97d6f | 239 | |
emh203 | 0:3d9c67d97d6f | 240 | /* Decrement the column loop counter */ |
emh203 | 0:3d9c67d97d6f | 241 | col--; |
emh203 | 0:3d9c67d97d6f | 242 | } |
emh203 | 0:3d9c67d97d6f | 243 | |
emh203 | 0:3d9c67d97d6f | 244 | /* If the columns of pSrcB is not a multiple of 4, compute any remaining output samples here. |
emh203 | 0:3d9c67d97d6f | 245 | ** No loop unrolling is used. */ |
emh203 | 0:3d9c67d97d6f | 246 | col = numColsB % 0x4u; |
emh203 | 0:3d9c67d97d6f | 247 | |
emh203 | 0:3d9c67d97d6f | 248 | while(col > 0u) |
emh203 | 0:3d9c67d97d6f | 249 | { |
emh203 | 0:3d9c67d97d6f | 250 | /* Read and store the input element in the destination */ |
emh203 | 0:3d9c67d97d6f | 251 | *px = *pInB++; |
emh203 | 0:3d9c67d97d6f | 252 | |
emh203 | 0:3d9c67d97d6f | 253 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emh203 | 0:3d9c67d97d6f | 254 | px += numRowsB; |
emh203 | 0:3d9c67d97d6f | 255 | |
emh203 | 0:3d9c67d97d6f | 256 | /* Decrement the column loop counter */ |
emh203 | 0:3d9c67d97d6f | 257 | col--; |
emh203 | 0:3d9c67d97d6f | 258 | } |
emh203 | 0:3d9c67d97d6f | 259 | |
emh203 | 0:3d9c67d97d6f | 260 | i++; |
emh203 | 0:3d9c67d97d6f | 261 | |
emh203 | 0:3d9c67d97d6f | 262 | /* Decrement the row loop counter */ |
emh203 | 0:3d9c67d97d6f | 263 | row--; |
emh203 | 0:3d9c67d97d6f | 264 | |
emh203 | 0:3d9c67d97d6f | 265 | } while(row > 0u); |
emh203 | 0:3d9c67d97d6f | 266 | |
emh203 | 0:3d9c67d97d6f | 267 | /* Reset the variables for the usage in the following multiplication process */ |
emh203 | 0:3d9c67d97d6f | 268 | row = numRowsA; |
emh203 | 0:3d9c67d97d6f | 269 | i = 0u; |
emh203 | 0:3d9c67d97d6f | 270 | px = pDst->pData; |
emh203 | 0:3d9c67d97d6f | 271 | |
emh203 | 0:3d9c67d97d6f | 272 | /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */ |
emh203 | 0:3d9c67d97d6f | 273 | /* row loop */ |
emh203 | 0:3d9c67d97d6f | 274 | do |
emh203 | 0:3d9c67d97d6f | 275 | { |
emh203 | 0:3d9c67d97d6f | 276 | /* For every row wise process, the column loop counter is to be initiated */ |
emh203 | 0:3d9c67d97d6f | 277 | col = numColsB; |
emh203 | 0:3d9c67d97d6f | 278 | |
emh203 | 0:3d9c67d97d6f | 279 | /* For every row wise process, the pIn2 pointer is set |
emh203 | 0:3d9c67d97d6f | 280 | ** to the starting address of the transposed pSrcB data */ |
emh203 | 0:3d9c67d97d6f | 281 | pInB = pSrcBT; |
emh203 | 0:3d9c67d97d6f | 282 | |
emh203 | 0:3d9c67d97d6f | 283 | /* column loop */ |
emh203 | 0:3d9c67d97d6f | 284 | do |
emh203 | 0:3d9c67d97d6f | 285 | { |
emh203 | 0:3d9c67d97d6f | 286 | /* Set the variable sum, that acts as accumulator, to zero */ |
emh203 | 0:3d9c67d97d6f | 287 | sum = 0; |
emh203 | 0:3d9c67d97d6f | 288 | |
emh203 | 0:3d9c67d97d6f | 289 | /* Apply loop unrolling and compute 2 MACs simultaneously. */ |
emh203 | 0:3d9c67d97d6f | 290 | colCnt = numColsA >> 2; |
emh203 | 0:3d9c67d97d6f | 291 | |
emh203 | 0:3d9c67d97d6f | 292 | /* Initiate the pointer pIn1 to point to the starting address of the column being processed */ |
emh203 | 0:3d9c67d97d6f | 293 | pInA = pSrcA->pData + i; |
emh203 | 0:3d9c67d97d6f | 294 | |
emh203 | 0:3d9c67d97d6f | 295 | /* matrix multiplication */ |
emh203 | 0:3d9c67d97d6f | 296 | while(colCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 297 | { |
emh203 | 0:3d9c67d97d6f | 298 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
emh203 | 0:3d9c67d97d6f | 299 | #ifndef UNALIGNED_SUPPORT_DISABLE |
emh203 | 0:3d9c67d97d6f | 300 | |
emh203 | 0:3d9c67d97d6f | 301 | inA1 = *__SIMD32(pInA)++; |
emh203 | 0:3d9c67d97d6f | 302 | inB1 = *__SIMD32(pInB)++; |
emh203 | 0:3d9c67d97d6f | 303 | inA2 = *__SIMD32(pInA)++; |
emh203 | 0:3d9c67d97d6f | 304 | inB2 = *__SIMD32(pInB)++; |
emh203 | 0:3d9c67d97d6f | 305 | |
emh203 | 0:3d9c67d97d6f | 306 | sum = __SMLAD(inA1, inB1, sum); |
emh203 | 0:3d9c67d97d6f | 307 | sum = __SMLAD(inA2, inB2, sum); |
emh203 | 0:3d9c67d97d6f | 308 | |
emh203 | 0:3d9c67d97d6f | 309 | #else |
emh203 | 0:3d9c67d97d6f | 310 | |
emh203 | 0:3d9c67d97d6f | 311 | inA1 = *pInA++; |
emh203 | 0:3d9c67d97d6f | 312 | inB1 = *pInB++; |
emh203 | 0:3d9c67d97d6f | 313 | inA2 = *pInA++; |
emh203 | 0:3d9c67d97d6f | 314 | sum += inA1 * inB1; |
emh203 | 0:3d9c67d97d6f | 315 | inB2 = *pInB++; |
emh203 | 0:3d9c67d97d6f | 316 | |
emh203 | 0:3d9c67d97d6f | 317 | inA1 = *pInA++; |
emh203 | 0:3d9c67d97d6f | 318 | inB1 = *pInB++; |
emh203 | 0:3d9c67d97d6f | 319 | sum += inA2 * inB2; |
emh203 | 0:3d9c67d97d6f | 320 | inA2 = *pInA++; |
emh203 | 0:3d9c67d97d6f | 321 | inB2 = *pInB++; |
emh203 | 0:3d9c67d97d6f | 322 | |
emh203 | 0:3d9c67d97d6f | 323 | sum += inA1 * inB1; |
emh203 | 0:3d9c67d97d6f | 324 | sum += inA2 * inB2; |
emh203 | 0:3d9c67d97d6f | 325 | |
emh203 | 0:3d9c67d97d6f | 326 | #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */ |
emh203 | 0:3d9c67d97d6f | 327 | |
emh203 | 0:3d9c67d97d6f | 328 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 329 | colCnt--; |
emh203 | 0:3d9c67d97d6f | 330 | } |
emh203 | 0:3d9c67d97d6f | 331 | |
emh203 | 0:3d9c67d97d6f | 332 | /* process odd column samples */ |
emh203 | 0:3d9c67d97d6f | 333 | colCnt = numColsA % 0x4u; |
emh203 | 0:3d9c67d97d6f | 334 | |
emh203 | 0:3d9c67d97d6f | 335 | while(colCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 336 | { |
emh203 | 0:3d9c67d97d6f | 337 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
emh203 | 0:3d9c67d97d6f | 338 | sum += (q31_t) (*pInA++) * (*pInB++); |
emh203 | 0:3d9c67d97d6f | 339 | |
emh203 | 0:3d9c67d97d6f | 340 | colCnt--; |
emh203 | 0:3d9c67d97d6f | 341 | } |
emh203 | 0:3d9c67d97d6f | 342 | |
emh203 | 0:3d9c67d97d6f | 343 | /* Saturate and store the result in the destination buffer */ |
emh203 | 0:3d9c67d97d6f | 344 | *px = (q15_t) (sum >> 15); |
emh203 | 0:3d9c67d97d6f | 345 | px++; |
emh203 | 0:3d9c67d97d6f | 346 | |
emh203 | 0:3d9c67d97d6f | 347 | /* Decrement the column loop counter */ |
emh203 | 0:3d9c67d97d6f | 348 | col--; |
emh203 | 0:3d9c67d97d6f | 349 | |
emh203 | 0:3d9c67d97d6f | 350 | } while(col > 0u); |
emh203 | 0:3d9c67d97d6f | 351 | |
emh203 | 0:3d9c67d97d6f | 352 | i = i + numColsA; |
emh203 | 0:3d9c67d97d6f | 353 | |
emh203 | 0:3d9c67d97d6f | 354 | /* Decrement the row loop counter */ |
emh203 | 0:3d9c67d97d6f | 355 | row--; |
emh203 | 0:3d9c67d97d6f | 356 | |
emh203 | 0:3d9c67d97d6f | 357 | } while(row > 0u); |
emh203 | 0:3d9c67d97d6f | 358 | |
emh203 | 0:3d9c67d97d6f | 359 | /* set status as ARM_MATH_SUCCESS */ |
emh203 | 0:3d9c67d97d6f | 360 | status = ARM_MATH_SUCCESS; |
emh203 | 0:3d9c67d97d6f | 361 | } |
emh203 | 0:3d9c67d97d6f | 362 | |
emh203 | 0:3d9c67d97d6f | 363 | /* Return to application */ |
emh203 | 0:3d9c67d97d6f | 364 | return (status); |
emh203 | 0:3d9c67d97d6f | 365 | } |
emh203 | 0:3d9c67d97d6f | 366 | |
emh203 | 0:3d9c67d97d6f | 367 | /** |
emh203 | 0:3d9c67d97d6f | 368 | * @} end of MatrixMult group |
emh203 | 0:3d9c67d97d6f | 369 | */ |