Fork of mbed-dsp. CMSIS-DSP library of supporting NEON
Dependents: mbed-os-example-cmsis_dsp_neon
Fork of mbed-dsp by
Information
Japanese version is available in lower part of this page.
このページの後半に日本語版が用意されています.
CMSIS-DSP of supporting NEON
What is this ?
A library for CMSIS-DSP of supporting NEON.
We supported the NEON to CMSIS-DSP Ver1.4.3(CMSIS V4.1) that ARM supplied, has achieved the processing speed improvement.
If you use the mbed-dsp library, you can use to replace this library.
CMSIS-DSP of supporting NEON is provied as a library.
Library Creation environment
CMSIS-DSP library of supporting NEON was created by the following environment.
- Compiler
ARMCC Version 5.03 - Compile option switch[C Compiler]
-DARM_MATH_MATRIX_CHECK -DARM_MATH_ROUNDING -O3 -Otime --cpu=Cortex-A9 --littleend --arm --apcs=/interwork --no_unaligned_access --fpu=vfpv3_fp16 --fpmode=fast --apcs=/hardfp --vectorize --asm
- Compile option switch[Assembler]
--cpreproc --cpu=Cortex-A9 --littleend --arm --apcs=/interwork --no_unaligned_access --fpu=vfpv3_fp16 --fpmode=fast --apcs=/hardfp
Effects of NEON support
In the data which passes to each function, large size will be expected more effective than small size.
Also if the data is a multiple of 16, effect will be expected in every function in the CMSIS-DSP.
NEON対応CMSIS-DSP
概要
NEON対応したCMSIS-DSPのライブラリです。
ARM社提供のCMSIS-DSP Ver1.4.3(CMSIS V4.1)をターゲットにNEON対応を行ない、処理速度向上を実現しております。
mbed-dspライブラリを使用している場合は、本ライブラリに置き換えて使用することができます。
NEON対応したCMSIS-DSPはライブラリで提供します。
ライブラリ作成環境
NEON対応CMSIS-DSPライブラリは、以下の環境で作成しています。
- コンパイラ
ARMCC Version 5.03 - コンパイルオプションスイッチ[C Compiler]
-DARM_MATH_MATRIX_CHECK -DARM_MATH_ROUNDING -O3 -Otime --cpu=Cortex-A9 --littleend --arm --apcs=/interwork --no_unaligned_access --fpu=vfpv3_fp16 --fpmode=fast --apcs=/hardfp --vectorize --asm
- コンパイルオプションスイッチ[Assembler]
--cpreproc --cpu=Cortex-A9 --littleend --arm --apcs=/interwork --no_unaligned_access --fpu=vfpv3_fp16 --fpmode=fast --apcs=/hardfp
NEON対応による効果について
CMSIS-DSP内の各関数へ渡すデータは、小さいサイズよりも大きいサイズの方が効果が見込めます。
また、16の倍数のデータであれば、CMSIS-DSP内のどの関数でも効果が見込めます。
cmsis_dsp/MatrixFunctions/arm_mat_mult_q15.c@1:fdd22bb7aa52, 2012-11-28 (annotated)
- Committer:
- emilmont
- Date:
- Wed Nov 28 12:30:09 2012 +0000
- Revision:
- 1:fdd22bb7aa52
- Child:
- 2:da51fb522205
DSP library code
Who changed what in which revision?
User | Revision | Line number | New contents of line |
---|---|---|---|
emilmont | 1:fdd22bb7aa52 | 1 | /* ---------------------------------------------------------------------- |
emilmont | 1:fdd22bb7aa52 | 2 | * Copyright (C) 2010 ARM Limited. All rights reserved. |
emilmont | 1:fdd22bb7aa52 | 3 | * |
emilmont | 1:fdd22bb7aa52 | 4 | * $Date: 15. February 2012 |
emilmont | 1:fdd22bb7aa52 | 5 | * $Revision: V1.1.0 |
emilmont | 1:fdd22bb7aa52 | 6 | * |
emilmont | 1:fdd22bb7aa52 | 7 | * Project: CMSIS DSP Library |
emilmont | 1:fdd22bb7aa52 | 8 | * Title: arm_mat_mult_q15.c |
emilmont | 1:fdd22bb7aa52 | 9 | * |
emilmont | 1:fdd22bb7aa52 | 10 | * Description: Q15 matrix multiplication. |
emilmont | 1:fdd22bb7aa52 | 11 | * |
emilmont | 1:fdd22bb7aa52 | 12 | * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0 |
emilmont | 1:fdd22bb7aa52 | 13 | * |
emilmont | 1:fdd22bb7aa52 | 14 | * Version 1.1.0 2012/02/15 |
emilmont | 1:fdd22bb7aa52 | 15 | * Updated with more optimizations, bug fixes and minor API changes. |
emilmont | 1:fdd22bb7aa52 | 16 | * |
emilmont | 1:fdd22bb7aa52 | 17 | * Version 1.0.10 2011/7/15 |
emilmont | 1:fdd22bb7aa52 | 18 | * Big Endian support added and Merged M0 and M3/M4 Source code. |
emilmont | 1:fdd22bb7aa52 | 19 | * |
emilmont | 1:fdd22bb7aa52 | 20 | * Version 1.0.3 2010/11/29 |
emilmont | 1:fdd22bb7aa52 | 21 | * Re-organized the CMSIS folders and updated documentation. |
emilmont | 1:fdd22bb7aa52 | 22 | * |
emilmont | 1:fdd22bb7aa52 | 23 | * Version 1.0.2 2010/11/11 |
emilmont | 1:fdd22bb7aa52 | 24 | * Documentation updated. |
emilmont | 1:fdd22bb7aa52 | 25 | * |
emilmont | 1:fdd22bb7aa52 | 26 | * Version 1.0.1 2010/10/05 |
emilmont | 1:fdd22bb7aa52 | 27 | * Production release and review comments incorporated. |
emilmont | 1:fdd22bb7aa52 | 28 | * |
emilmont | 1:fdd22bb7aa52 | 29 | * Version 1.0.0 2010/09/20 |
emilmont | 1:fdd22bb7aa52 | 30 | * Production release and review comments incorporated. |
emilmont | 1:fdd22bb7aa52 | 31 | * |
emilmont | 1:fdd22bb7aa52 | 32 | * Version 0.0.5 2010/04/26 |
emilmont | 1:fdd22bb7aa52 | 33 | * incorporated review comments and updated with latest CMSIS layer |
emilmont | 1:fdd22bb7aa52 | 34 | * |
emilmont | 1:fdd22bb7aa52 | 35 | * Version 0.0.3 2010/03/10 |
emilmont | 1:fdd22bb7aa52 | 36 | * Initial version |
emilmont | 1:fdd22bb7aa52 | 37 | * -------------------------------------------------------------------- */ |
emilmont | 1:fdd22bb7aa52 | 38 | |
emilmont | 1:fdd22bb7aa52 | 39 | #include "arm_math.h" |
emilmont | 1:fdd22bb7aa52 | 40 | |
emilmont | 1:fdd22bb7aa52 | 41 | /** |
emilmont | 1:fdd22bb7aa52 | 42 | * @ingroup groupMatrix |
emilmont | 1:fdd22bb7aa52 | 43 | */ |
emilmont | 1:fdd22bb7aa52 | 44 | |
emilmont | 1:fdd22bb7aa52 | 45 | /** |
emilmont | 1:fdd22bb7aa52 | 46 | * @addtogroup MatrixMult |
emilmont | 1:fdd22bb7aa52 | 47 | * @{ |
emilmont | 1:fdd22bb7aa52 | 48 | */ |
emilmont | 1:fdd22bb7aa52 | 49 | |
emilmont | 1:fdd22bb7aa52 | 50 | |
emilmont | 1:fdd22bb7aa52 | 51 | /** |
emilmont | 1:fdd22bb7aa52 | 52 | * @brief Q15 matrix multiplication |
emilmont | 1:fdd22bb7aa52 | 53 | * @param[in] *pSrcA points to the first input matrix structure |
emilmont | 1:fdd22bb7aa52 | 54 | * @param[in] *pSrcB points to the second input matrix structure |
emilmont | 1:fdd22bb7aa52 | 55 | * @param[out] *pDst points to output matrix structure |
emilmont | 1:fdd22bb7aa52 | 56 | * @param[in] *pState points to the array for storing intermediate results |
emilmont | 1:fdd22bb7aa52 | 57 | * @return The function returns either |
emilmont | 1:fdd22bb7aa52 | 58 | * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking. |
emilmont | 1:fdd22bb7aa52 | 59 | * |
emilmont | 1:fdd22bb7aa52 | 60 | * @details |
emilmont | 1:fdd22bb7aa52 | 61 | * <b>Scaling and Overflow Behavior:</b> |
emilmont | 1:fdd22bb7aa52 | 62 | * |
emilmont | 1:fdd22bb7aa52 | 63 | * \par |
emilmont | 1:fdd22bb7aa52 | 64 | * The function is implemented using a 64-bit internal accumulator. The inputs to the |
emilmont | 1:fdd22bb7aa52 | 65 | * multiplications are in 1.15 format and multiplications yield a 2.30 result. |
emilmont | 1:fdd22bb7aa52 | 66 | * The 2.30 intermediate |
emilmont | 1:fdd22bb7aa52 | 67 | * results are accumulated in a 64-bit accumulator in 34.30 format. This approach |
emilmont | 1:fdd22bb7aa52 | 68 | * provides 33 guard bits and there is no risk of overflow. The 34.30 result is then |
emilmont | 1:fdd22bb7aa52 | 69 | * truncated to 34.15 format by discarding the low 15 bits and then saturated to |
emilmont | 1:fdd22bb7aa52 | 70 | * 1.15 format. |
emilmont | 1:fdd22bb7aa52 | 71 | * |
emilmont | 1:fdd22bb7aa52 | 72 | * \par |
emilmont | 1:fdd22bb7aa52 | 73 | * 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. |
emilmont | 1:fdd22bb7aa52 | 74 | * |
emilmont | 1:fdd22bb7aa52 | 75 | */ |
emilmont | 1:fdd22bb7aa52 | 76 | |
emilmont | 1:fdd22bb7aa52 | 77 | arm_status arm_mat_mult_q15( |
emilmont | 1:fdd22bb7aa52 | 78 | const arm_matrix_instance_q15 * pSrcA, |
emilmont | 1:fdd22bb7aa52 | 79 | const arm_matrix_instance_q15 * pSrcB, |
emilmont | 1:fdd22bb7aa52 | 80 | arm_matrix_instance_q15 * pDst, |
emilmont | 1:fdd22bb7aa52 | 81 | q15_t * pState) |
emilmont | 1:fdd22bb7aa52 | 82 | { |
emilmont | 1:fdd22bb7aa52 | 83 | q63_t sum; /* accumulator */ |
emilmont | 1:fdd22bb7aa52 | 84 | |
emilmont | 1:fdd22bb7aa52 | 85 | #ifndef ARM_MATH_CM0 |
emilmont | 1:fdd22bb7aa52 | 86 | |
emilmont | 1:fdd22bb7aa52 | 87 | /* Run the below code for Cortex-M4 and Cortex-M3 */ |
emilmont | 1:fdd22bb7aa52 | 88 | |
emilmont | 1:fdd22bb7aa52 | 89 | q15_t *pSrcBT = pState; /* input data matrix pointer for transpose */ |
emilmont | 1:fdd22bb7aa52 | 90 | q15_t *pInA = pSrcA->pData; /* input data matrix pointer A of Q15 type */ |
emilmont | 1:fdd22bb7aa52 | 91 | q15_t *pInB = pSrcB->pData; /* input data matrix pointer B of Q15 type */ |
emilmont | 1:fdd22bb7aa52 | 92 | q15_t *px; /* Temporary output data matrix pointer */ |
emilmont | 1:fdd22bb7aa52 | 93 | uint16_t numRowsA = pSrcA->numRows; /* number of rows of input matrix A */ |
emilmont | 1:fdd22bb7aa52 | 94 | uint16_t numColsB = pSrcB->numCols; /* number of columns of input matrix B */ |
emilmont | 1:fdd22bb7aa52 | 95 | uint16_t numColsA = pSrcA->numCols; /* number of columns of input matrix A */ |
emilmont | 1:fdd22bb7aa52 | 96 | uint16_t numRowsB = pSrcB->numRows; /* number of rows of input matrix A */ |
emilmont | 1:fdd22bb7aa52 | 97 | uint16_t col, i = 0u, row = numRowsB, colCnt; /* loop counters */ |
emilmont | 1:fdd22bb7aa52 | 98 | arm_status status; /* status of matrix multiplication */ |
emilmont | 1:fdd22bb7aa52 | 99 | |
emilmont | 1:fdd22bb7aa52 | 100 | #ifndef UNALIGNED_SUPPORT_DISABLE |
emilmont | 1:fdd22bb7aa52 | 101 | |
emilmont | 1:fdd22bb7aa52 | 102 | q31_t in; /* Temporary variable to hold the input value */ |
emilmont | 1:fdd22bb7aa52 | 103 | q31_t pSourceA1, pSourceB1, pSourceA2, pSourceB2; |
emilmont | 1:fdd22bb7aa52 | 104 | |
emilmont | 1:fdd22bb7aa52 | 105 | #else |
emilmont | 1:fdd22bb7aa52 | 106 | |
emilmont | 1:fdd22bb7aa52 | 107 | q15_t in; /* Temporary variable to hold the input value */ |
emilmont | 1:fdd22bb7aa52 | 108 | q15_t inA1, inB1, inA2, inB2; |
emilmont | 1:fdd22bb7aa52 | 109 | |
emilmont | 1:fdd22bb7aa52 | 110 | #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */ |
emilmont | 1:fdd22bb7aa52 | 111 | |
emilmont | 1:fdd22bb7aa52 | 112 | #ifdef ARM_MATH_MATRIX_CHECK |
emilmont | 1:fdd22bb7aa52 | 113 | /* Check for matrix mismatch condition */ |
emilmont | 1:fdd22bb7aa52 | 114 | if((pSrcA->numCols != pSrcB->numRows) || |
emilmont | 1:fdd22bb7aa52 | 115 | (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols)) |
emilmont | 1:fdd22bb7aa52 | 116 | { |
emilmont | 1:fdd22bb7aa52 | 117 | /* Set status as ARM_MATH_SIZE_MISMATCH */ |
emilmont | 1:fdd22bb7aa52 | 118 | status = ARM_MATH_SIZE_MISMATCH; |
emilmont | 1:fdd22bb7aa52 | 119 | } |
emilmont | 1:fdd22bb7aa52 | 120 | else |
emilmont | 1:fdd22bb7aa52 | 121 | #endif /* #ifdef ARM_MATH_MATRIX_CHECK */ |
emilmont | 1:fdd22bb7aa52 | 122 | { |
emilmont | 1:fdd22bb7aa52 | 123 | /* Matrix transpose */ |
emilmont | 1:fdd22bb7aa52 | 124 | do |
emilmont | 1:fdd22bb7aa52 | 125 | { |
emilmont | 1:fdd22bb7aa52 | 126 | /* Apply loop unrolling and exchange the columns with row elements */ |
emilmont | 1:fdd22bb7aa52 | 127 | col = numColsB >> 2; |
emilmont | 1:fdd22bb7aa52 | 128 | |
emilmont | 1:fdd22bb7aa52 | 129 | /* The pointer px is set to starting address of the column being processed */ |
emilmont | 1:fdd22bb7aa52 | 130 | px = pSrcBT + i; |
emilmont | 1:fdd22bb7aa52 | 131 | |
emilmont | 1:fdd22bb7aa52 | 132 | /* First part of the processing with loop unrolling. Compute 4 outputs at a time. |
emilmont | 1:fdd22bb7aa52 | 133 | ** a second loop below computes the remaining 1 to 3 samples. */ |
emilmont | 1:fdd22bb7aa52 | 134 | while(col > 0u) |
emilmont | 1:fdd22bb7aa52 | 135 | { |
emilmont | 1:fdd22bb7aa52 | 136 | #ifndef UNALIGNED_SUPPORT_DISABLE |
emilmont | 1:fdd22bb7aa52 | 137 | |
emilmont | 1:fdd22bb7aa52 | 138 | /* Read two elements from the row */ |
emilmont | 1:fdd22bb7aa52 | 139 | in = *__SIMD32(pInB)++; |
emilmont | 1:fdd22bb7aa52 | 140 | |
emilmont | 1:fdd22bb7aa52 | 141 | /* Unpack and store one element in the destination */ |
emilmont | 1:fdd22bb7aa52 | 142 | #ifndef ARM_MATH_BIG_ENDIAN |
emilmont | 1:fdd22bb7aa52 | 143 | |
emilmont | 1:fdd22bb7aa52 | 144 | *px = (q15_t) in; |
emilmont | 1:fdd22bb7aa52 | 145 | |
emilmont | 1:fdd22bb7aa52 | 146 | #else |
emilmont | 1:fdd22bb7aa52 | 147 | |
emilmont | 1:fdd22bb7aa52 | 148 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
emilmont | 1:fdd22bb7aa52 | 149 | |
emilmont | 1:fdd22bb7aa52 | 150 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
emilmont | 1:fdd22bb7aa52 | 151 | |
emilmont | 1:fdd22bb7aa52 | 152 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emilmont | 1:fdd22bb7aa52 | 153 | px += numRowsB; |
emilmont | 1:fdd22bb7aa52 | 154 | |
emilmont | 1:fdd22bb7aa52 | 155 | /* Unpack and store the second element in the destination */ |
emilmont | 1:fdd22bb7aa52 | 156 | #ifndef ARM_MATH_BIG_ENDIAN |
emilmont | 1:fdd22bb7aa52 | 157 | |
emilmont | 1:fdd22bb7aa52 | 158 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
emilmont | 1:fdd22bb7aa52 | 159 | |
emilmont | 1:fdd22bb7aa52 | 160 | #else |
emilmont | 1:fdd22bb7aa52 | 161 | |
emilmont | 1:fdd22bb7aa52 | 162 | *px = (q15_t) in; |
emilmont | 1:fdd22bb7aa52 | 163 | |
emilmont | 1:fdd22bb7aa52 | 164 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
emilmont | 1:fdd22bb7aa52 | 165 | |
emilmont | 1:fdd22bb7aa52 | 166 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emilmont | 1:fdd22bb7aa52 | 167 | px += numRowsB; |
emilmont | 1:fdd22bb7aa52 | 168 | |
emilmont | 1:fdd22bb7aa52 | 169 | /* Read two elements from the row */ |
emilmont | 1:fdd22bb7aa52 | 170 | in = *__SIMD32(pInB)++; |
emilmont | 1:fdd22bb7aa52 | 171 | |
emilmont | 1:fdd22bb7aa52 | 172 | /* Unpack and store one element in the destination */ |
emilmont | 1:fdd22bb7aa52 | 173 | #ifndef ARM_MATH_BIG_ENDIAN |
emilmont | 1:fdd22bb7aa52 | 174 | |
emilmont | 1:fdd22bb7aa52 | 175 | *px = (q15_t) in; |
emilmont | 1:fdd22bb7aa52 | 176 | |
emilmont | 1:fdd22bb7aa52 | 177 | #else |
emilmont | 1:fdd22bb7aa52 | 178 | |
emilmont | 1:fdd22bb7aa52 | 179 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
emilmont | 1:fdd22bb7aa52 | 180 | |
emilmont | 1:fdd22bb7aa52 | 181 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
emilmont | 1:fdd22bb7aa52 | 182 | |
emilmont | 1:fdd22bb7aa52 | 183 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emilmont | 1:fdd22bb7aa52 | 184 | px += numRowsB; |
emilmont | 1:fdd22bb7aa52 | 185 | |
emilmont | 1:fdd22bb7aa52 | 186 | /* Unpack and store the second element in the destination */ |
emilmont | 1:fdd22bb7aa52 | 187 | |
emilmont | 1:fdd22bb7aa52 | 188 | #ifndef ARM_MATH_BIG_ENDIAN |
emilmont | 1:fdd22bb7aa52 | 189 | |
emilmont | 1:fdd22bb7aa52 | 190 | *px = (q15_t) ((in & (q31_t) 0xffff0000) >> 16); |
emilmont | 1:fdd22bb7aa52 | 191 | |
emilmont | 1:fdd22bb7aa52 | 192 | #else |
emilmont | 1:fdd22bb7aa52 | 193 | |
emilmont | 1:fdd22bb7aa52 | 194 | *px = (q15_t) in; |
emilmont | 1:fdd22bb7aa52 | 195 | |
emilmont | 1:fdd22bb7aa52 | 196 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
emilmont | 1:fdd22bb7aa52 | 197 | |
emilmont | 1:fdd22bb7aa52 | 198 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emilmont | 1:fdd22bb7aa52 | 199 | px += numRowsB; |
emilmont | 1:fdd22bb7aa52 | 200 | |
emilmont | 1:fdd22bb7aa52 | 201 | #else |
emilmont | 1:fdd22bb7aa52 | 202 | |
emilmont | 1:fdd22bb7aa52 | 203 | /* Read one element from the row */ |
emilmont | 1:fdd22bb7aa52 | 204 | in = *pInB++; |
emilmont | 1:fdd22bb7aa52 | 205 | |
emilmont | 1:fdd22bb7aa52 | 206 | /* Store one element in the destination */ |
emilmont | 1:fdd22bb7aa52 | 207 | *px = in; |
emilmont | 1:fdd22bb7aa52 | 208 | |
emilmont | 1:fdd22bb7aa52 | 209 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emilmont | 1:fdd22bb7aa52 | 210 | px += numRowsB; |
emilmont | 1:fdd22bb7aa52 | 211 | |
emilmont | 1:fdd22bb7aa52 | 212 | /* Read one element from the row */ |
emilmont | 1:fdd22bb7aa52 | 213 | in = *pInB++; |
emilmont | 1:fdd22bb7aa52 | 214 | |
emilmont | 1:fdd22bb7aa52 | 215 | /* Store one element in the destination */ |
emilmont | 1:fdd22bb7aa52 | 216 | *px = in; |
emilmont | 1:fdd22bb7aa52 | 217 | |
emilmont | 1:fdd22bb7aa52 | 218 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emilmont | 1:fdd22bb7aa52 | 219 | px += numRowsB; |
emilmont | 1:fdd22bb7aa52 | 220 | |
emilmont | 1:fdd22bb7aa52 | 221 | /* Read one element from the row */ |
emilmont | 1:fdd22bb7aa52 | 222 | in = *pInB++; |
emilmont | 1:fdd22bb7aa52 | 223 | |
emilmont | 1:fdd22bb7aa52 | 224 | /* Store one element in the destination */ |
emilmont | 1:fdd22bb7aa52 | 225 | *px = in; |
emilmont | 1:fdd22bb7aa52 | 226 | |
emilmont | 1:fdd22bb7aa52 | 227 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emilmont | 1:fdd22bb7aa52 | 228 | px += numRowsB; |
emilmont | 1:fdd22bb7aa52 | 229 | |
emilmont | 1:fdd22bb7aa52 | 230 | /* Read one element from the row */ |
emilmont | 1:fdd22bb7aa52 | 231 | in = *pInB++; |
emilmont | 1:fdd22bb7aa52 | 232 | |
emilmont | 1:fdd22bb7aa52 | 233 | /* Store one element in the destination */ |
emilmont | 1:fdd22bb7aa52 | 234 | *px = in; |
emilmont | 1:fdd22bb7aa52 | 235 | |
emilmont | 1:fdd22bb7aa52 | 236 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emilmont | 1:fdd22bb7aa52 | 237 | px += numRowsB; |
emilmont | 1:fdd22bb7aa52 | 238 | |
emilmont | 1:fdd22bb7aa52 | 239 | #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */ |
emilmont | 1:fdd22bb7aa52 | 240 | |
emilmont | 1:fdd22bb7aa52 | 241 | /* Decrement the column loop counter */ |
emilmont | 1:fdd22bb7aa52 | 242 | col--; |
emilmont | 1:fdd22bb7aa52 | 243 | } |
emilmont | 1:fdd22bb7aa52 | 244 | |
emilmont | 1:fdd22bb7aa52 | 245 | /* If the columns of pSrcB is not a multiple of 4, compute any remaining output samples here. |
emilmont | 1:fdd22bb7aa52 | 246 | ** No loop unrolling is used. */ |
emilmont | 1:fdd22bb7aa52 | 247 | col = numColsB % 0x4u; |
emilmont | 1:fdd22bb7aa52 | 248 | |
emilmont | 1:fdd22bb7aa52 | 249 | while(col > 0u) |
emilmont | 1:fdd22bb7aa52 | 250 | { |
emilmont | 1:fdd22bb7aa52 | 251 | /* Read and store the input element in the destination */ |
emilmont | 1:fdd22bb7aa52 | 252 | *px = *pInB++; |
emilmont | 1:fdd22bb7aa52 | 253 | |
emilmont | 1:fdd22bb7aa52 | 254 | /* Update the pointer px to point to the next row of the transposed matrix */ |
emilmont | 1:fdd22bb7aa52 | 255 | px += numRowsB; |
emilmont | 1:fdd22bb7aa52 | 256 | |
emilmont | 1:fdd22bb7aa52 | 257 | /* Decrement the column loop counter */ |
emilmont | 1:fdd22bb7aa52 | 258 | col--; |
emilmont | 1:fdd22bb7aa52 | 259 | } |
emilmont | 1:fdd22bb7aa52 | 260 | |
emilmont | 1:fdd22bb7aa52 | 261 | i++; |
emilmont | 1:fdd22bb7aa52 | 262 | |
emilmont | 1:fdd22bb7aa52 | 263 | /* Decrement the row loop counter */ |
emilmont | 1:fdd22bb7aa52 | 264 | row--; |
emilmont | 1:fdd22bb7aa52 | 265 | |
emilmont | 1:fdd22bb7aa52 | 266 | } while(row > 0u); |
emilmont | 1:fdd22bb7aa52 | 267 | |
emilmont | 1:fdd22bb7aa52 | 268 | /* Reset the variables for the usage in the following multiplication process */ |
emilmont | 1:fdd22bb7aa52 | 269 | row = numRowsA; |
emilmont | 1:fdd22bb7aa52 | 270 | i = 0u; |
emilmont | 1:fdd22bb7aa52 | 271 | px = pDst->pData; |
emilmont | 1:fdd22bb7aa52 | 272 | |
emilmont | 1:fdd22bb7aa52 | 273 | /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */ |
emilmont | 1:fdd22bb7aa52 | 274 | /* row loop */ |
emilmont | 1:fdd22bb7aa52 | 275 | do |
emilmont | 1:fdd22bb7aa52 | 276 | { |
emilmont | 1:fdd22bb7aa52 | 277 | /* For every row wise process, the column loop counter is to be initiated */ |
emilmont | 1:fdd22bb7aa52 | 278 | col = numColsB; |
emilmont | 1:fdd22bb7aa52 | 279 | |
emilmont | 1:fdd22bb7aa52 | 280 | /* For every row wise process, the pIn2 pointer is set |
emilmont | 1:fdd22bb7aa52 | 281 | ** to the starting address of the transposed pSrcB data */ |
emilmont | 1:fdd22bb7aa52 | 282 | pInB = pSrcBT; |
emilmont | 1:fdd22bb7aa52 | 283 | |
emilmont | 1:fdd22bb7aa52 | 284 | /* column loop */ |
emilmont | 1:fdd22bb7aa52 | 285 | do |
emilmont | 1:fdd22bb7aa52 | 286 | { |
emilmont | 1:fdd22bb7aa52 | 287 | /* Set the variable sum, that acts as accumulator, to zero */ |
emilmont | 1:fdd22bb7aa52 | 288 | sum = 0; |
emilmont | 1:fdd22bb7aa52 | 289 | |
emilmont | 1:fdd22bb7aa52 | 290 | /* Apply loop unrolling and compute 2 MACs simultaneously. */ |
emilmont | 1:fdd22bb7aa52 | 291 | colCnt = numColsA >> 2; |
emilmont | 1:fdd22bb7aa52 | 292 | |
emilmont | 1:fdd22bb7aa52 | 293 | /* Initiate the pointer pIn1 to point to the starting address of the column being processed */ |
emilmont | 1:fdd22bb7aa52 | 294 | pInA = pSrcA->pData + i; |
emilmont | 1:fdd22bb7aa52 | 295 | |
emilmont | 1:fdd22bb7aa52 | 296 | |
emilmont | 1:fdd22bb7aa52 | 297 | /* matrix multiplication */ |
emilmont | 1:fdd22bb7aa52 | 298 | while(colCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 299 | { |
emilmont | 1:fdd22bb7aa52 | 300 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
emilmont | 1:fdd22bb7aa52 | 301 | #ifndef UNALIGNED_SUPPORT_DISABLE |
emilmont | 1:fdd22bb7aa52 | 302 | |
emilmont | 1:fdd22bb7aa52 | 303 | /* read real and imag values from pSrcA and pSrcB buffer */ |
emilmont | 1:fdd22bb7aa52 | 304 | pSourceA1 = *__SIMD32(pInA)++; |
emilmont | 1:fdd22bb7aa52 | 305 | pSourceB1 = *__SIMD32(pInB)++; |
emilmont | 1:fdd22bb7aa52 | 306 | |
emilmont | 1:fdd22bb7aa52 | 307 | pSourceA2 = *__SIMD32(pInA)++; |
emilmont | 1:fdd22bb7aa52 | 308 | pSourceB2 = *__SIMD32(pInB)++; |
emilmont | 1:fdd22bb7aa52 | 309 | |
emilmont | 1:fdd22bb7aa52 | 310 | /* Multiply and Accumlates */ |
emilmont | 1:fdd22bb7aa52 | 311 | sum = __SMLALD(pSourceA1, pSourceB1, sum); |
emilmont | 1:fdd22bb7aa52 | 312 | sum = __SMLALD(pSourceA2, pSourceB2, sum); |
emilmont | 1:fdd22bb7aa52 | 313 | |
emilmont | 1:fdd22bb7aa52 | 314 | #else |
emilmont | 1:fdd22bb7aa52 | 315 | /* read real and imag values from pSrcA and pSrcB buffer */ |
emilmont | 1:fdd22bb7aa52 | 316 | inA1 = *pInA++; |
emilmont | 1:fdd22bb7aa52 | 317 | inB1 = *pInB++; |
emilmont | 1:fdd22bb7aa52 | 318 | inA2 = *pInA++; |
emilmont | 1:fdd22bb7aa52 | 319 | /* Multiply and Accumlates */ |
emilmont | 1:fdd22bb7aa52 | 320 | sum += inA1 * inB1; |
emilmont | 1:fdd22bb7aa52 | 321 | inB2 = *pInB++; |
emilmont | 1:fdd22bb7aa52 | 322 | |
emilmont | 1:fdd22bb7aa52 | 323 | inA1 = *pInA++; |
emilmont | 1:fdd22bb7aa52 | 324 | inB1 = *pInB++; |
emilmont | 1:fdd22bb7aa52 | 325 | /* Multiply and Accumlates */ |
emilmont | 1:fdd22bb7aa52 | 326 | sum += inA2 * inB2; |
emilmont | 1:fdd22bb7aa52 | 327 | inA2 = *pInA++; |
emilmont | 1:fdd22bb7aa52 | 328 | inB2 = *pInB++; |
emilmont | 1:fdd22bb7aa52 | 329 | |
emilmont | 1:fdd22bb7aa52 | 330 | /* Multiply and Accumlates */ |
emilmont | 1:fdd22bb7aa52 | 331 | sum += inA1 * inB1; |
emilmont | 1:fdd22bb7aa52 | 332 | sum += inA2 * inB2; |
emilmont | 1:fdd22bb7aa52 | 333 | |
emilmont | 1:fdd22bb7aa52 | 334 | #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */ |
emilmont | 1:fdd22bb7aa52 | 335 | |
emilmont | 1:fdd22bb7aa52 | 336 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 337 | colCnt--; |
emilmont | 1:fdd22bb7aa52 | 338 | } |
emilmont | 1:fdd22bb7aa52 | 339 | |
emilmont | 1:fdd22bb7aa52 | 340 | /* process remaining column samples */ |
emilmont | 1:fdd22bb7aa52 | 341 | colCnt = numColsA & 3u; |
emilmont | 1:fdd22bb7aa52 | 342 | |
emilmont | 1:fdd22bb7aa52 | 343 | while(colCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 344 | { |
emilmont | 1:fdd22bb7aa52 | 345 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
emilmont | 1:fdd22bb7aa52 | 346 | sum += *pInA++ * *pInB++; |
emilmont | 1:fdd22bb7aa52 | 347 | |
emilmont | 1:fdd22bb7aa52 | 348 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 349 | colCnt--; |
emilmont | 1:fdd22bb7aa52 | 350 | } |
emilmont | 1:fdd22bb7aa52 | 351 | |
emilmont | 1:fdd22bb7aa52 | 352 | /* Saturate and store the result in the destination buffer */ |
emilmont | 1:fdd22bb7aa52 | 353 | *px = (q15_t) (__SSAT((sum >> 15), 16)); |
emilmont | 1:fdd22bb7aa52 | 354 | px++; |
emilmont | 1:fdd22bb7aa52 | 355 | |
emilmont | 1:fdd22bb7aa52 | 356 | /* Decrement the column loop counter */ |
emilmont | 1:fdd22bb7aa52 | 357 | col--; |
emilmont | 1:fdd22bb7aa52 | 358 | |
emilmont | 1:fdd22bb7aa52 | 359 | } while(col > 0u); |
emilmont | 1:fdd22bb7aa52 | 360 | |
emilmont | 1:fdd22bb7aa52 | 361 | i = i + numColsA; |
emilmont | 1:fdd22bb7aa52 | 362 | |
emilmont | 1:fdd22bb7aa52 | 363 | /* Decrement the row loop counter */ |
emilmont | 1:fdd22bb7aa52 | 364 | row--; |
emilmont | 1:fdd22bb7aa52 | 365 | |
emilmont | 1:fdd22bb7aa52 | 366 | } while(row > 0u); |
emilmont | 1:fdd22bb7aa52 | 367 | |
emilmont | 1:fdd22bb7aa52 | 368 | #else |
emilmont | 1:fdd22bb7aa52 | 369 | |
emilmont | 1:fdd22bb7aa52 | 370 | /* Run the below code for Cortex-M0 */ |
emilmont | 1:fdd22bb7aa52 | 371 | |
emilmont | 1:fdd22bb7aa52 | 372 | q15_t *pIn1 = pSrcA->pData; /* input data matrix pointer A */ |
emilmont | 1:fdd22bb7aa52 | 373 | q15_t *pIn2 = pSrcB->pData; /* input data matrix pointer B */ |
emilmont | 1:fdd22bb7aa52 | 374 | q15_t *pInA = pSrcA->pData; /* input data matrix pointer A of Q15 type */ |
emilmont | 1:fdd22bb7aa52 | 375 | q15_t *pInB = pSrcB->pData; /* input data matrix pointer B of Q15 type */ |
emilmont | 1:fdd22bb7aa52 | 376 | q15_t *pOut = pDst->pData; /* output data matrix pointer */ |
emilmont | 1:fdd22bb7aa52 | 377 | q15_t *px; /* Temporary output data matrix pointer */ |
emilmont | 1:fdd22bb7aa52 | 378 | uint16_t numColsB = pSrcB->numCols; /* number of columns of input matrix B */ |
emilmont | 1:fdd22bb7aa52 | 379 | uint16_t numColsA = pSrcA->numCols; /* number of columns of input matrix A */ |
emilmont | 1:fdd22bb7aa52 | 380 | uint16_t numRowsA = pSrcA->numRows; /* number of rows of input matrix A */ |
emilmont | 1:fdd22bb7aa52 | 381 | uint16_t col, i = 0u, row = numRowsA, colCnt; /* loop counters */ |
emilmont | 1:fdd22bb7aa52 | 382 | arm_status status; /* status of matrix multiplication */ |
emilmont | 1:fdd22bb7aa52 | 383 | |
emilmont | 1:fdd22bb7aa52 | 384 | #ifdef ARM_MATH_MATRIX_CHECK |
emilmont | 1:fdd22bb7aa52 | 385 | |
emilmont | 1:fdd22bb7aa52 | 386 | /* Check for matrix mismatch condition */ |
emilmont | 1:fdd22bb7aa52 | 387 | if((pSrcA->numCols != pSrcB->numRows) || |
emilmont | 1:fdd22bb7aa52 | 388 | (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols)) |
emilmont | 1:fdd22bb7aa52 | 389 | { |
emilmont | 1:fdd22bb7aa52 | 390 | /* Set status as ARM_MATH_SIZE_MISMATCH */ |
emilmont | 1:fdd22bb7aa52 | 391 | status = ARM_MATH_SIZE_MISMATCH; |
emilmont | 1:fdd22bb7aa52 | 392 | } |
emilmont | 1:fdd22bb7aa52 | 393 | else |
emilmont | 1:fdd22bb7aa52 | 394 | #endif /* #ifdef ARM_MATH_MATRIX_CHECK */ |
emilmont | 1:fdd22bb7aa52 | 395 | |
emilmont | 1:fdd22bb7aa52 | 396 | { |
emilmont | 1:fdd22bb7aa52 | 397 | /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */ |
emilmont | 1:fdd22bb7aa52 | 398 | /* row loop */ |
emilmont | 1:fdd22bb7aa52 | 399 | do |
emilmont | 1:fdd22bb7aa52 | 400 | { |
emilmont | 1:fdd22bb7aa52 | 401 | /* Output pointer is set to starting address of the row being processed */ |
emilmont | 1:fdd22bb7aa52 | 402 | px = pOut + i; |
emilmont | 1:fdd22bb7aa52 | 403 | |
emilmont | 1:fdd22bb7aa52 | 404 | /* For every row wise process, the column loop counter is to be initiated */ |
emilmont | 1:fdd22bb7aa52 | 405 | col = numColsB; |
emilmont | 1:fdd22bb7aa52 | 406 | |
emilmont | 1:fdd22bb7aa52 | 407 | /* For every row wise process, the pIn2 pointer is set |
emilmont | 1:fdd22bb7aa52 | 408 | ** to the starting address of the pSrcB data */ |
emilmont | 1:fdd22bb7aa52 | 409 | pIn2 = pSrcB->pData; |
emilmont | 1:fdd22bb7aa52 | 410 | |
emilmont | 1:fdd22bb7aa52 | 411 | /* column loop */ |
emilmont | 1:fdd22bb7aa52 | 412 | do |
emilmont | 1:fdd22bb7aa52 | 413 | { |
emilmont | 1:fdd22bb7aa52 | 414 | /* Set the variable sum, that acts as accumulator, to zero */ |
emilmont | 1:fdd22bb7aa52 | 415 | sum = 0; |
emilmont | 1:fdd22bb7aa52 | 416 | |
emilmont | 1:fdd22bb7aa52 | 417 | /* Initiate the pointer pIn1 to point to the starting address of pSrcA */ |
emilmont | 1:fdd22bb7aa52 | 418 | pIn1 = pInA; |
emilmont | 1:fdd22bb7aa52 | 419 | |
emilmont | 1:fdd22bb7aa52 | 420 | /* Matrix A columns number of MAC operations are to be performed */ |
emilmont | 1:fdd22bb7aa52 | 421 | colCnt = numColsA; |
emilmont | 1:fdd22bb7aa52 | 422 | |
emilmont | 1:fdd22bb7aa52 | 423 | /* matrix multiplication */ |
emilmont | 1:fdd22bb7aa52 | 424 | while(colCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 425 | { |
emilmont | 1:fdd22bb7aa52 | 426 | /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
emilmont | 1:fdd22bb7aa52 | 427 | /* Perform the multiply-accumulates */ |
emilmont | 1:fdd22bb7aa52 | 428 | sum += (q31_t) * pIn1++ * *pIn2; |
emilmont | 1:fdd22bb7aa52 | 429 | pIn2 += numColsB; |
emilmont | 1:fdd22bb7aa52 | 430 | |
emilmont | 1:fdd22bb7aa52 | 431 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 432 | colCnt--; |
emilmont | 1:fdd22bb7aa52 | 433 | } |
emilmont | 1:fdd22bb7aa52 | 434 | |
emilmont | 1:fdd22bb7aa52 | 435 | /* Convert the result from 34.30 to 1.15 format and store the saturated value in destination buffer */ |
emilmont | 1:fdd22bb7aa52 | 436 | /* Saturate and store the result in the destination buffer */ |
emilmont | 1:fdd22bb7aa52 | 437 | *px++ = (q15_t) __SSAT((sum >> 15), 16); |
emilmont | 1:fdd22bb7aa52 | 438 | |
emilmont | 1:fdd22bb7aa52 | 439 | /* Decrement the column loop counter */ |
emilmont | 1:fdd22bb7aa52 | 440 | col--; |
emilmont | 1:fdd22bb7aa52 | 441 | |
emilmont | 1:fdd22bb7aa52 | 442 | /* Update the pointer pIn2 to point to the starting address of the next column */ |
emilmont | 1:fdd22bb7aa52 | 443 | pIn2 = pInB + (numColsB - col); |
emilmont | 1:fdd22bb7aa52 | 444 | |
emilmont | 1:fdd22bb7aa52 | 445 | } while(col > 0u); |
emilmont | 1:fdd22bb7aa52 | 446 | |
emilmont | 1:fdd22bb7aa52 | 447 | /* Update the pointer pSrcA to point to the starting address of the next row */ |
emilmont | 1:fdd22bb7aa52 | 448 | i = i + numColsB; |
emilmont | 1:fdd22bb7aa52 | 449 | pInA = pInA + numColsA; |
emilmont | 1:fdd22bb7aa52 | 450 | |
emilmont | 1:fdd22bb7aa52 | 451 | /* Decrement the row loop counter */ |
emilmont | 1:fdd22bb7aa52 | 452 | row--; |
emilmont | 1:fdd22bb7aa52 | 453 | |
emilmont | 1:fdd22bb7aa52 | 454 | } while(row > 0u); |
emilmont | 1:fdd22bb7aa52 | 455 | |
emilmont | 1:fdd22bb7aa52 | 456 | #endif /* #ifndef ARM_MATH_CM0 */ |
emilmont | 1:fdd22bb7aa52 | 457 | /* set status as ARM_MATH_SUCCESS */ |
emilmont | 1:fdd22bb7aa52 | 458 | status = ARM_MATH_SUCCESS; |
emilmont | 1:fdd22bb7aa52 | 459 | } |
emilmont | 1:fdd22bb7aa52 | 460 | |
emilmont | 1:fdd22bb7aa52 | 461 | /* Return to application */ |
emilmont | 1:fdd22bb7aa52 | 462 | return (status); |
emilmont | 1:fdd22bb7aa52 | 463 | } |
emilmont | 1:fdd22bb7aa52 | 464 | |
emilmont | 1:fdd22bb7aa52 | 465 | /** |
emilmont | 1:fdd22bb7aa52 | 466 | * @} end of MatrixMult group |
emilmont | 1:fdd22bb7aa52 | 467 | */ |