CMSIS DSP library
Dependents: performance_timer Surfboard_ gps2rtty Capstone ... more
Legacy Warning
This is an mbed 2 library. To learn more about mbed OS 5, visit the docs.
cmsis_dsp/FilteringFunctions/arm_conv_partial_fast_opt_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_conv_partial_fast_opt_q15.c |
emilmont | 1:fdd22bb7aa52 | 9 | * |
emilmont | 1:fdd22bb7aa52 | 10 | * Description: Fast Q15 Partial convolution. |
emilmont | 1:fdd22bb7aa52 | 11 | * |
emilmont | 1:fdd22bb7aa52 | 12 | * Target Processor: Cortex-M4/Cortex-M3 |
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 1.0.11 2011/10/18 |
emilmont | 1:fdd22bb7aa52 | 18 | * Bug Fix in conv, correlation, partial convolution. |
emilmont | 1:fdd22bb7aa52 | 19 | * |
emilmont | 1:fdd22bb7aa52 | 20 | * Version 1.0.10 2011/7/15 |
emilmont | 1:fdd22bb7aa52 | 21 | * Big Endian support added and Merged M0 and M3/M4 Source code. |
emilmont | 1:fdd22bb7aa52 | 22 | * |
emilmont | 1:fdd22bb7aa52 | 23 | * Version 1.0.3 2010/11/29 |
emilmont | 1:fdd22bb7aa52 | 24 | * Re-organized the CMSIS folders and updated documentation. |
emilmont | 1:fdd22bb7aa52 | 25 | * |
emilmont | 1:fdd22bb7aa52 | 26 | * Version 1.0.2 2010/11/11 |
emilmont | 1:fdd22bb7aa52 | 27 | * Documentation updated. |
emilmont | 1:fdd22bb7aa52 | 28 | * |
emilmont | 1:fdd22bb7aa52 | 29 | * Version 1.0.1 2010/10/05 |
emilmont | 1:fdd22bb7aa52 | 30 | * Production release and review comments incorporated. |
emilmont | 1:fdd22bb7aa52 | 31 | * |
emilmont | 1:fdd22bb7aa52 | 32 | * Version 1.0.0 2010/09/20 |
emilmont | 1:fdd22bb7aa52 | 33 | * Production release and review comments incorporated. |
emilmont | 1:fdd22bb7aa52 | 34 | * -------------------------------------------------------------------- */ |
emilmont | 1:fdd22bb7aa52 | 35 | |
emilmont | 1:fdd22bb7aa52 | 36 | #include "arm_math.h" |
emilmont | 1:fdd22bb7aa52 | 37 | |
emilmont | 1:fdd22bb7aa52 | 38 | /** |
emilmont | 1:fdd22bb7aa52 | 39 | * @ingroup groupFilters |
emilmont | 1:fdd22bb7aa52 | 40 | */ |
emilmont | 1:fdd22bb7aa52 | 41 | |
emilmont | 1:fdd22bb7aa52 | 42 | /** |
emilmont | 1:fdd22bb7aa52 | 43 | * @addtogroup PartialConv |
emilmont | 1:fdd22bb7aa52 | 44 | * @{ |
emilmont | 1:fdd22bb7aa52 | 45 | */ |
emilmont | 1:fdd22bb7aa52 | 46 | |
emilmont | 1:fdd22bb7aa52 | 47 | /** |
emilmont | 1:fdd22bb7aa52 | 48 | * @brief Partial convolution of Q15 sequences (fast version) for Cortex-M3 and Cortex-M4. |
emilmont | 1:fdd22bb7aa52 | 49 | * @param[in] *pSrcA points to the first input sequence. |
emilmont | 1:fdd22bb7aa52 | 50 | * @param[in] srcALen length of the first input sequence. |
emilmont | 1:fdd22bb7aa52 | 51 | * @param[in] *pSrcB points to the second input sequence. |
emilmont | 1:fdd22bb7aa52 | 52 | * @param[in] srcBLen length of the second input sequence. |
emilmont | 1:fdd22bb7aa52 | 53 | * @param[out] *pDst points to the location where the output result is written. |
emilmont | 1:fdd22bb7aa52 | 54 | * @param[in] firstIndex is the first output sample to start with. |
emilmont | 1:fdd22bb7aa52 | 55 | * @param[in] numPoints is the number of output points to be computed. |
emilmont | 1:fdd22bb7aa52 | 56 | * @param[in] *pScratch1 points to scratch buffer of size max(srcALen, srcBLen) + 2*min(srcALen, srcBLen) - 2. |
emilmont | 1:fdd22bb7aa52 | 57 | * @param[in] *pScratch2 points to scratch buffer of size min(srcALen, srcBLen). |
emilmont | 1:fdd22bb7aa52 | 58 | * @return Returns either ARM_MATH_SUCCESS if the function completed correctly or ARM_MATH_ARGUMENT_ERROR if the requested subset is not in the range [0 srcALen+srcBLen-2]. |
emilmont | 1:fdd22bb7aa52 | 59 | * |
emilmont | 1:fdd22bb7aa52 | 60 | * See <code>arm_conv_partial_q15()</code> for a slower implementation of this function which uses a 64-bit accumulator to avoid wrap around distortion. |
emilmont | 1:fdd22bb7aa52 | 61 | * |
emilmont | 1:fdd22bb7aa52 | 62 | * \par Restrictions |
emilmont | 1:fdd22bb7aa52 | 63 | * If the silicon does not support unaligned memory access enable the macro UNALIGNED_SUPPORT_DISABLE |
emilmont | 1:fdd22bb7aa52 | 64 | * In this case input, output, scratch1 and scratch2 buffers should be aligned by 32-bit |
emilmont | 1:fdd22bb7aa52 | 65 | * |
emilmont | 1:fdd22bb7aa52 | 66 | */ |
emilmont | 1:fdd22bb7aa52 | 67 | |
emilmont | 1:fdd22bb7aa52 | 68 | #ifndef UNALIGNED_SUPPORT_DISABLE |
emilmont | 1:fdd22bb7aa52 | 69 | |
emilmont | 1:fdd22bb7aa52 | 70 | arm_status arm_conv_partial_fast_opt_q15( |
emilmont | 1:fdd22bb7aa52 | 71 | q15_t * pSrcA, |
emilmont | 1:fdd22bb7aa52 | 72 | uint32_t srcALen, |
emilmont | 1:fdd22bb7aa52 | 73 | q15_t * pSrcB, |
emilmont | 1:fdd22bb7aa52 | 74 | uint32_t srcBLen, |
emilmont | 1:fdd22bb7aa52 | 75 | q15_t * pDst, |
emilmont | 1:fdd22bb7aa52 | 76 | uint32_t firstIndex, |
emilmont | 1:fdd22bb7aa52 | 77 | uint32_t numPoints, |
emilmont | 1:fdd22bb7aa52 | 78 | q15_t * pScratch1, |
emilmont | 1:fdd22bb7aa52 | 79 | q15_t * pScratch2) |
emilmont | 1:fdd22bb7aa52 | 80 | { |
emilmont | 1:fdd22bb7aa52 | 81 | |
emilmont | 1:fdd22bb7aa52 | 82 | q15_t *pOut = pDst; /* output pointer */ |
emilmont | 1:fdd22bb7aa52 | 83 | q15_t *pScr1 = pScratch1; /* Temporary pointer for scratch1 */ |
emilmont | 1:fdd22bb7aa52 | 84 | q15_t *pScr2 = pScratch2; /* Temporary pointer for scratch1 */ |
emilmont | 1:fdd22bb7aa52 | 85 | q31_t acc0, acc1, acc2, acc3; /* Accumulator */ |
emilmont | 1:fdd22bb7aa52 | 86 | q31_t x1, x2, x3; /* Temporary variables to hold state and coefficient values */ |
emilmont | 1:fdd22bb7aa52 | 87 | q31_t y1, y2; /* State variables */ |
emilmont | 1:fdd22bb7aa52 | 88 | q15_t *pIn1; /* inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 89 | q15_t *pIn2; /* inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 90 | q15_t *px; /* Intermediate inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 91 | q15_t *py; /* Intermediate inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 92 | uint32_t j, k, blkCnt; /* loop counter */ |
emilmont | 1:fdd22bb7aa52 | 93 | arm_status status; |
emilmont | 1:fdd22bb7aa52 | 94 | |
emilmont | 1:fdd22bb7aa52 | 95 | uint32_t tapCnt; /* loop count */ |
emilmont | 1:fdd22bb7aa52 | 96 | |
emilmont | 1:fdd22bb7aa52 | 97 | /* Check for range of output samples to be calculated */ |
emilmont | 1:fdd22bb7aa52 | 98 | if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u)))) |
emilmont | 1:fdd22bb7aa52 | 99 | { |
emilmont | 1:fdd22bb7aa52 | 100 | /* Set status as ARM_MATH_ARGUMENT_ERROR */ |
emilmont | 1:fdd22bb7aa52 | 101 | status = ARM_MATH_ARGUMENT_ERROR; |
emilmont | 1:fdd22bb7aa52 | 102 | } |
emilmont | 1:fdd22bb7aa52 | 103 | else |
emilmont | 1:fdd22bb7aa52 | 104 | { |
emilmont | 1:fdd22bb7aa52 | 105 | |
emilmont | 1:fdd22bb7aa52 | 106 | /* The algorithm implementation is based on the lengths of the inputs. */ |
emilmont | 1:fdd22bb7aa52 | 107 | /* srcB is always made to slide across srcA. */ |
emilmont | 1:fdd22bb7aa52 | 108 | /* So srcBLen is always considered as shorter or equal to srcALen */ |
emilmont | 1:fdd22bb7aa52 | 109 | if(srcALen >= srcBLen) |
emilmont | 1:fdd22bb7aa52 | 110 | { |
emilmont | 1:fdd22bb7aa52 | 111 | /* Initialization of inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 112 | pIn1 = pSrcA; |
emilmont | 1:fdd22bb7aa52 | 113 | |
emilmont | 1:fdd22bb7aa52 | 114 | /* Initialization of inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 115 | pIn2 = pSrcB; |
emilmont | 1:fdd22bb7aa52 | 116 | } |
emilmont | 1:fdd22bb7aa52 | 117 | else |
emilmont | 1:fdd22bb7aa52 | 118 | { |
emilmont | 1:fdd22bb7aa52 | 119 | /* Initialization of inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 120 | pIn1 = pSrcB; |
emilmont | 1:fdd22bb7aa52 | 121 | |
emilmont | 1:fdd22bb7aa52 | 122 | /* Initialization of inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 123 | pIn2 = pSrcA; |
emilmont | 1:fdd22bb7aa52 | 124 | |
emilmont | 1:fdd22bb7aa52 | 125 | /* srcBLen is always considered as shorter or equal to srcALen */ |
emilmont | 1:fdd22bb7aa52 | 126 | j = srcBLen; |
emilmont | 1:fdd22bb7aa52 | 127 | srcBLen = srcALen; |
emilmont | 1:fdd22bb7aa52 | 128 | srcALen = j; |
emilmont | 1:fdd22bb7aa52 | 129 | } |
emilmont | 1:fdd22bb7aa52 | 130 | |
emilmont | 1:fdd22bb7aa52 | 131 | /* Temporary pointer for scratch2 */ |
emilmont | 1:fdd22bb7aa52 | 132 | py = pScratch2; |
emilmont | 1:fdd22bb7aa52 | 133 | |
emilmont | 1:fdd22bb7aa52 | 134 | /* pointer to take end of scratch2 buffer */ |
emilmont | 1:fdd22bb7aa52 | 135 | pScr2 = pScratch2 + srcBLen - 1; |
emilmont | 1:fdd22bb7aa52 | 136 | |
emilmont | 1:fdd22bb7aa52 | 137 | /* points to smaller length sequence */ |
emilmont | 1:fdd22bb7aa52 | 138 | px = pIn2; |
emilmont | 1:fdd22bb7aa52 | 139 | |
emilmont | 1:fdd22bb7aa52 | 140 | /* Apply loop unrolling and do 4 Copies simultaneously. */ |
emilmont | 1:fdd22bb7aa52 | 141 | k = srcBLen >> 2u; |
emilmont | 1:fdd22bb7aa52 | 142 | |
emilmont | 1:fdd22bb7aa52 | 143 | /* First part of the processing with loop unrolling copies 4 data points at a time. |
emilmont | 1:fdd22bb7aa52 | 144 | ** a second loop below copies for the remaining 1 to 3 samples. */ |
emilmont | 1:fdd22bb7aa52 | 145 | |
emilmont | 1:fdd22bb7aa52 | 146 | /* Copy smaller length input sequence in reverse order into second scratch buffer */ |
emilmont | 1:fdd22bb7aa52 | 147 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 148 | { |
emilmont | 1:fdd22bb7aa52 | 149 | /* copy second buffer in reversal manner */ |
emilmont | 1:fdd22bb7aa52 | 150 | *pScr2-- = *px++; |
emilmont | 1:fdd22bb7aa52 | 151 | *pScr2-- = *px++; |
emilmont | 1:fdd22bb7aa52 | 152 | *pScr2-- = *px++; |
emilmont | 1:fdd22bb7aa52 | 153 | *pScr2-- = *px++; |
emilmont | 1:fdd22bb7aa52 | 154 | |
emilmont | 1:fdd22bb7aa52 | 155 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 156 | k--; |
emilmont | 1:fdd22bb7aa52 | 157 | } |
emilmont | 1:fdd22bb7aa52 | 158 | |
emilmont | 1:fdd22bb7aa52 | 159 | /* If the count is not a multiple of 4, copy remaining samples here. |
emilmont | 1:fdd22bb7aa52 | 160 | ** No loop unrolling is used. */ |
emilmont | 1:fdd22bb7aa52 | 161 | k = srcBLen % 0x4u; |
emilmont | 1:fdd22bb7aa52 | 162 | |
emilmont | 1:fdd22bb7aa52 | 163 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 164 | { |
emilmont | 1:fdd22bb7aa52 | 165 | /* copy second buffer in reversal manner for remaining samples */ |
emilmont | 1:fdd22bb7aa52 | 166 | *pScr2-- = *px++; |
emilmont | 1:fdd22bb7aa52 | 167 | |
emilmont | 1:fdd22bb7aa52 | 168 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 169 | k--; |
emilmont | 1:fdd22bb7aa52 | 170 | } |
emilmont | 1:fdd22bb7aa52 | 171 | |
emilmont | 1:fdd22bb7aa52 | 172 | /* Initialze temporary scratch pointer */ |
emilmont | 1:fdd22bb7aa52 | 173 | pScr1 = pScratch1; |
emilmont | 1:fdd22bb7aa52 | 174 | |
emilmont | 1:fdd22bb7aa52 | 175 | /* Assuming scratch1 buffer is aligned by 32-bit */ |
emilmont | 1:fdd22bb7aa52 | 176 | /* Fill (srcBLen - 1u) zeros in scratch buffer */ |
emilmont | 1:fdd22bb7aa52 | 177 | arm_fill_q15(0, pScr1, (srcBLen - 1u)); |
emilmont | 1:fdd22bb7aa52 | 178 | |
emilmont | 1:fdd22bb7aa52 | 179 | /* Update temporary scratch pointer */ |
emilmont | 1:fdd22bb7aa52 | 180 | pScr1 += (srcBLen - 1u); |
emilmont | 1:fdd22bb7aa52 | 181 | |
emilmont | 1:fdd22bb7aa52 | 182 | /* Copy bigger length sequence(srcALen) samples in scratch1 buffer */ |
emilmont | 1:fdd22bb7aa52 | 183 | |
emilmont | 1:fdd22bb7aa52 | 184 | /* Copy (srcALen) samples in scratch buffer */ |
emilmont | 1:fdd22bb7aa52 | 185 | arm_copy_q15(pIn1, pScr1, srcALen); |
emilmont | 1:fdd22bb7aa52 | 186 | |
emilmont | 1:fdd22bb7aa52 | 187 | /* Update pointers */ |
emilmont | 1:fdd22bb7aa52 | 188 | pScr1 += srcALen; |
emilmont | 1:fdd22bb7aa52 | 189 | |
emilmont | 1:fdd22bb7aa52 | 190 | /* Fill (srcBLen - 1u) zeros at end of scratch buffer */ |
emilmont | 1:fdd22bb7aa52 | 191 | arm_fill_q15(0, pScr1, (srcBLen - 1u)); |
emilmont | 1:fdd22bb7aa52 | 192 | |
emilmont | 1:fdd22bb7aa52 | 193 | /* Update pointer */ |
emilmont | 1:fdd22bb7aa52 | 194 | pScr1 += (srcBLen - 1u); |
emilmont | 1:fdd22bb7aa52 | 195 | |
emilmont | 1:fdd22bb7aa52 | 196 | /* Initialization of pIn2 pointer */ |
emilmont | 1:fdd22bb7aa52 | 197 | pIn2 = py; |
emilmont | 1:fdd22bb7aa52 | 198 | |
emilmont | 1:fdd22bb7aa52 | 199 | pScratch1 += firstIndex; |
emilmont | 1:fdd22bb7aa52 | 200 | |
emilmont | 1:fdd22bb7aa52 | 201 | pOut = pDst + firstIndex; |
emilmont | 1:fdd22bb7aa52 | 202 | |
emilmont | 1:fdd22bb7aa52 | 203 | /* First part of the processing with loop unrolling process 4 data points at a time. |
emilmont | 1:fdd22bb7aa52 | 204 | ** a second loop below process for the remaining 1 to 3 samples. */ |
emilmont | 1:fdd22bb7aa52 | 205 | |
emilmont | 1:fdd22bb7aa52 | 206 | /* Actual convolution process starts here */ |
emilmont | 1:fdd22bb7aa52 | 207 | blkCnt = (numPoints) >> 2; |
emilmont | 1:fdd22bb7aa52 | 208 | |
emilmont | 1:fdd22bb7aa52 | 209 | while(blkCnt > 0) |
emilmont | 1:fdd22bb7aa52 | 210 | { |
emilmont | 1:fdd22bb7aa52 | 211 | /* Initialze temporary scratch pointer as scratch1 */ |
emilmont | 1:fdd22bb7aa52 | 212 | pScr1 = pScratch1; |
emilmont | 1:fdd22bb7aa52 | 213 | |
emilmont | 1:fdd22bb7aa52 | 214 | /* Clear Accumlators */ |
emilmont | 1:fdd22bb7aa52 | 215 | acc0 = 0; |
emilmont | 1:fdd22bb7aa52 | 216 | acc1 = 0; |
emilmont | 1:fdd22bb7aa52 | 217 | acc2 = 0; |
emilmont | 1:fdd22bb7aa52 | 218 | acc3 = 0; |
emilmont | 1:fdd22bb7aa52 | 219 | |
emilmont | 1:fdd22bb7aa52 | 220 | /* Read two samples from scratch1 buffer */ |
emilmont | 1:fdd22bb7aa52 | 221 | x1 = *__SIMD32(pScr1)++; |
emilmont | 1:fdd22bb7aa52 | 222 | |
emilmont | 1:fdd22bb7aa52 | 223 | /* Read next two samples from scratch1 buffer */ |
emilmont | 1:fdd22bb7aa52 | 224 | x2 = *__SIMD32(pScr1)++; |
emilmont | 1:fdd22bb7aa52 | 225 | |
emilmont | 1:fdd22bb7aa52 | 226 | tapCnt = (srcBLen) >> 2u; |
emilmont | 1:fdd22bb7aa52 | 227 | |
emilmont | 1:fdd22bb7aa52 | 228 | while(tapCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 229 | { |
emilmont | 1:fdd22bb7aa52 | 230 | |
emilmont | 1:fdd22bb7aa52 | 231 | /* Read four samples from smaller buffer */ |
emilmont | 1:fdd22bb7aa52 | 232 | y1 = _SIMD32_OFFSET(pIn2); |
emilmont | 1:fdd22bb7aa52 | 233 | y2 = _SIMD32_OFFSET(pIn2 + 2u); |
emilmont | 1:fdd22bb7aa52 | 234 | |
emilmont | 1:fdd22bb7aa52 | 235 | /* multiply and accumlate */ |
emilmont | 1:fdd22bb7aa52 | 236 | acc0 = __SMLAD(x1, y1, acc0); |
emilmont | 1:fdd22bb7aa52 | 237 | acc2 = __SMLAD(x2, y1, acc2); |
emilmont | 1:fdd22bb7aa52 | 238 | |
emilmont | 1:fdd22bb7aa52 | 239 | /* pack input data */ |
emilmont | 1:fdd22bb7aa52 | 240 | #ifndef ARM_MATH_BIG_ENDIAN |
emilmont | 1:fdd22bb7aa52 | 241 | x3 = __PKHBT(x2, x1, 0); |
emilmont | 1:fdd22bb7aa52 | 242 | #else |
emilmont | 1:fdd22bb7aa52 | 243 | x3 = __PKHBT(x1, x2, 0); |
emilmont | 1:fdd22bb7aa52 | 244 | #endif |
emilmont | 1:fdd22bb7aa52 | 245 | |
emilmont | 1:fdd22bb7aa52 | 246 | /* multiply and accumlate */ |
emilmont | 1:fdd22bb7aa52 | 247 | acc1 = __SMLADX(x3, y1, acc1); |
emilmont | 1:fdd22bb7aa52 | 248 | |
emilmont | 1:fdd22bb7aa52 | 249 | /* Read next two samples from scratch1 buffer */ |
emilmont | 1:fdd22bb7aa52 | 250 | x1 = _SIMD32_OFFSET(pScr1); |
emilmont | 1:fdd22bb7aa52 | 251 | |
emilmont | 1:fdd22bb7aa52 | 252 | /* multiply and accumlate */ |
emilmont | 1:fdd22bb7aa52 | 253 | acc0 = __SMLAD(x2, y2, acc0); |
emilmont | 1:fdd22bb7aa52 | 254 | |
emilmont | 1:fdd22bb7aa52 | 255 | acc2 = __SMLAD(x1, y2, acc2); |
emilmont | 1:fdd22bb7aa52 | 256 | |
emilmont | 1:fdd22bb7aa52 | 257 | /* pack input data */ |
emilmont | 1:fdd22bb7aa52 | 258 | #ifndef ARM_MATH_BIG_ENDIAN |
emilmont | 1:fdd22bb7aa52 | 259 | x3 = __PKHBT(x1, x2, 0); |
emilmont | 1:fdd22bb7aa52 | 260 | #else |
emilmont | 1:fdd22bb7aa52 | 261 | x3 = __PKHBT(x2, x1, 0); |
emilmont | 1:fdd22bb7aa52 | 262 | #endif |
emilmont | 1:fdd22bb7aa52 | 263 | |
emilmont | 1:fdd22bb7aa52 | 264 | acc3 = __SMLADX(x3, y1, acc3); |
emilmont | 1:fdd22bb7aa52 | 265 | acc1 = __SMLADX(x3, y2, acc1); |
emilmont | 1:fdd22bb7aa52 | 266 | |
emilmont | 1:fdd22bb7aa52 | 267 | x2 = _SIMD32_OFFSET(pScr1 + 2u); |
emilmont | 1:fdd22bb7aa52 | 268 | |
emilmont | 1:fdd22bb7aa52 | 269 | #ifndef ARM_MATH_BIG_ENDIAN |
emilmont | 1:fdd22bb7aa52 | 270 | x3 = __PKHBT(x2, x1, 0); |
emilmont | 1:fdd22bb7aa52 | 271 | #else |
emilmont | 1:fdd22bb7aa52 | 272 | x3 = __PKHBT(x1, x2, 0); |
emilmont | 1:fdd22bb7aa52 | 273 | #endif |
emilmont | 1:fdd22bb7aa52 | 274 | |
emilmont | 1:fdd22bb7aa52 | 275 | acc3 = __SMLADX(x3, y2, acc3); |
emilmont | 1:fdd22bb7aa52 | 276 | |
emilmont | 1:fdd22bb7aa52 | 277 | /* update scratch pointers */ |
emilmont | 1:fdd22bb7aa52 | 278 | pIn2 += 4u; |
emilmont | 1:fdd22bb7aa52 | 279 | pScr1 += 4u; |
emilmont | 1:fdd22bb7aa52 | 280 | |
emilmont | 1:fdd22bb7aa52 | 281 | |
emilmont | 1:fdd22bb7aa52 | 282 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 283 | tapCnt--; |
emilmont | 1:fdd22bb7aa52 | 284 | } |
emilmont | 1:fdd22bb7aa52 | 285 | |
emilmont | 1:fdd22bb7aa52 | 286 | /* Update scratch pointer for remaining samples of smaller length sequence */ |
emilmont | 1:fdd22bb7aa52 | 287 | pScr1 -= 4u; |
emilmont | 1:fdd22bb7aa52 | 288 | |
emilmont | 1:fdd22bb7aa52 | 289 | /* apply same above for remaining samples of smaller length sequence */ |
emilmont | 1:fdd22bb7aa52 | 290 | tapCnt = (srcBLen) & 3u; |
emilmont | 1:fdd22bb7aa52 | 291 | |
emilmont | 1:fdd22bb7aa52 | 292 | while(tapCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 293 | { |
emilmont | 1:fdd22bb7aa52 | 294 | |
emilmont | 1:fdd22bb7aa52 | 295 | /* accumlate the results */ |
emilmont | 1:fdd22bb7aa52 | 296 | acc0 += (*pScr1++ * *pIn2); |
emilmont | 1:fdd22bb7aa52 | 297 | acc1 += (*pScr1++ * *pIn2); |
emilmont | 1:fdd22bb7aa52 | 298 | acc2 += (*pScr1++ * *pIn2); |
emilmont | 1:fdd22bb7aa52 | 299 | acc3 += (*pScr1++ * *pIn2++); |
emilmont | 1:fdd22bb7aa52 | 300 | |
emilmont | 1:fdd22bb7aa52 | 301 | pScr1 -= 3u; |
emilmont | 1:fdd22bb7aa52 | 302 | |
emilmont | 1:fdd22bb7aa52 | 303 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 304 | tapCnt--; |
emilmont | 1:fdd22bb7aa52 | 305 | } |
emilmont | 1:fdd22bb7aa52 | 306 | |
emilmont | 1:fdd22bb7aa52 | 307 | blkCnt--; |
emilmont | 1:fdd22bb7aa52 | 308 | |
emilmont | 1:fdd22bb7aa52 | 309 | |
emilmont | 1:fdd22bb7aa52 | 310 | /* Store the results in the accumulators in the destination buffer. */ |
emilmont | 1:fdd22bb7aa52 | 311 | |
emilmont | 1:fdd22bb7aa52 | 312 | #ifndef ARM_MATH_BIG_ENDIAN |
emilmont | 1:fdd22bb7aa52 | 313 | |
emilmont | 1:fdd22bb7aa52 | 314 | *__SIMD32(pOut)++ = |
emilmont | 1:fdd22bb7aa52 | 315 | __PKHBT(__SSAT((acc0 >> 15), 16), __SSAT((acc1 >> 15), 16), 16); |
emilmont | 1:fdd22bb7aa52 | 316 | *__SIMD32(pOut)++ = |
emilmont | 1:fdd22bb7aa52 | 317 | __PKHBT(__SSAT((acc2 >> 15), 16), __SSAT((acc3 >> 15), 16), 16); |
emilmont | 1:fdd22bb7aa52 | 318 | |
emilmont | 1:fdd22bb7aa52 | 319 | #else |
emilmont | 1:fdd22bb7aa52 | 320 | |
emilmont | 1:fdd22bb7aa52 | 321 | *__SIMD32(pOut)++ = |
emilmont | 1:fdd22bb7aa52 | 322 | __PKHBT(__SSAT((acc1 >> 15), 16), __SSAT((acc0 >> 15), 16), 16); |
emilmont | 1:fdd22bb7aa52 | 323 | *__SIMD32(pOut)++ = |
emilmont | 1:fdd22bb7aa52 | 324 | __PKHBT(__SSAT((acc3 >> 15), 16), __SSAT((acc2 >> 15), 16), 16); |
emilmont | 1:fdd22bb7aa52 | 325 | |
emilmont | 1:fdd22bb7aa52 | 326 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
emilmont | 1:fdd22bb7aa52 | 327 | |
emilmont | 1:fdd22bb7aa52 | 328 | /* Initialization of inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 329 | pIn2 = py; |
emilmont | 1:fdd22bb7aa52 | 330 | |
emilmont | 1:fdd22bb7aa52 | 331 | pScratch1 += 4u; |
emilmont | 1:fdd22bb7aa52 | 332 | |
emilmont | 1:fdd22bb7aa52 | 333 | } |
emilmont | 1:fdd22bb7aa52 | 334 | |
emilmont | 1:fdd22bb7aa52 | 335 | |
emilmont | 1:fdd22bb7aa52 | 336 | blkCnt = numPoints & 0x3; |
emilmont | 1:fdd22bb7aa52 | 337 | |
emilmont | 1:fdd22bb7aa52 | 338 | /* Calculate convolution for remaining samples of Bigger length sequence */ |
emilmont | 1:fdd22bb7aa52 | 339 | while(blkCnt > 0) |
emilmont | 1:fdd22bb7aa52 | 340 | { |
emilmont | 1:fdd22bb7aa52 | 341 | /* Initialze temporary scratch pointer as scratch1 */ |
emilmont | 1:fdd22bb7aa52 | 342 | pScr1 = pScratch1; |
emilmont | 1:fdd22bb7aa52 | 343 | |
emilmont | 1:fdd22bb7aa52 | 344 | /* Clear Accumlators */ |
emilmont | 1:fdd22bb7aa52 | 345 | acc0 = 0; |
emilmont | 1:fdd22bb7aa52 | 346 | |
emilmont | 1:fdd22bb7aa52 | 347 | tapCnt = (srcBLen) >> 1u; |
emilmont | 1:fdd22bb7aa52 | 348 | |
emilmont | 1:fdd22bb7aa52 | 349 | while(tapCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 350 | { |
emilmont | 1:fdd22bb7aa52 | 351 | |
emilmont | 1:fdd22bb7aa52 | 352 | /* Read next two samples from scratch1 buffer */ |
emilmont | 1:fdd22bb7aa52 | 353 | x1 = *__SIMD32(pScr1)++; |
emilmont | 1:fdd22bb7aa52 | 354 | |
emilmont | 1:fdd22bb7aa52 | 355 | /* Read two samples from smaller buffer */ |
emilmont | 1:fdd22bb7aa52 | 356 | y1 = *__SIMD32(pIn2)++; |
emilmont | 1:fdd22bb7aa52 | 357 | |
emilmont | 1:fdd22bb7aa52 | 358 | acc0 = __SMLAD(x1, y1, acc0); |
emilmont | 1:fdd22bb7aa52 | 359 | |
emilmont | 1:fdd22bb7aa52 | 360 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 361 | tapCnt--; |
emilmont | 1:fdd22bb7aa52 | 362 | } |
emilmont | 1:fdd22bb7aa52 | 363 | |
emilmont | 1:fdd22bb7aa52 | 364 | tapCnt = (srcBLen) & 1u; |
emilmont | 1:fdd22bb7aa52 | 365 | |
emilmont | 1:fdd22bb7aa52 | 366 | /* apply same above for remaining samples of smaller length sequence */ |
emilmont | 1:fdd22bb7aa52 | 367 | while(tapCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 368 | { |
emilmont | 1:fdd22bb7aa52 | 369 | |
emilmont | 1:fdd22bb7aa52 | 370 | /* accumlate the results */ |
emilmont | 1:fdd22bb7aa52 | 371 | acc0 += (*pScr1++ * *pIn2++); |
emilmont | 1:fdd22bb7aa52 | 372 | |
emilmont | 1:fdd22bb7aa52 | 373 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 374 | tapCnt--; |
emilmont | 1:fdd22bb7aa52 | 375 | } |
emilmont | 1:fdd22bb7aa52 | 376 | |
emilmont | 1:fdd22bb7aa52 | 377 | blkCnt--; |
emilmont | 1:fdd22bb7aa52 | 378 | |
emilmont | 1:fdd22bb7aa52 | 379 | /* The result is in 2.30 format. Convert to 1.15 with saturation. |
emilmont | 1:fdd22bb7aa52 | 380 | ** Then store the output in the destination buffer. */ |
emilmont | 1:fdd22bb7aa52 | 381 | *pOut++ = (q15_t) (__SSAT((acc0 >> 15), 16)); |
emilmont | 1:fdd22bb7aa52 | 382 | |
emilmont | 1:fdd22bb7aa52 | 383 | /* Initialization of inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 384 | pIn2 = py; |
emilmont | 1:fdd22bb7aa52 | 385 | |
emilmont | 1:fdd22bb7aa52 | 386 | pScratch1 += 1u; |
emilmont | 1:fdd22bb7aa52 | 387 | |
emilmont | 1:fdd22bb7aa52 | 388 | } |
emilmont | 1:fdd22bb7aa52 | 389 | /* set status as ARM_MATH_SUCCESS */ |
emilmont | 1:fdd22bb7aa52 | 390 | status = ARM_MATH_SUCCESS; |
emilmont | 1:fdd22bb7aa52 | 391 | } |
emilmont | 1:fdd22bb7aa52 | 392 | /* Return to application */ |
emilmont | 1:fdd22bb7aa52 | 393 | return (status); |
emilmont | 1:fdd22bb7aa52 | 394 | } |
emilmont | 1:fdd22bb7aa52 | 395 | |
emilmont | 1:fdd22bb7aa52 | 396 | #else |
emilmont | 1:fdd22bb7aa52 | 397 | |
emilmont | 1:fdd22bb7aa52 | 398 | arm_status arm_conv_partial_fast_opt_q15( |
emilmont | 1:fdd22bb7aa52 | 399 | q15_t * pSrcA, |
emilmont | 1:fdd22bb7aa52 | 400 | uint32_t srcALen, |
emilmont | 1:fdd22bb7aa52 | 401 | q15_t * pSrcB, |
emilmont | 1:fdd22bb7aa52 | 402 | uint32_t srcBLen, |
emilmont | 1:fdd22bb7aa52 | 403 | q15_t * pDst, |
emilmont | 1:fdd22bb7aa52 | 404 | uint32_t firstIndex, |
emilmont | 1:fdd22bb7aa52 | 405 | uint32_t numPoints, |
emilmont | 1:fdd22bb7aa52 | 406 | q15_t * pScratch1, |
emilmont | 1:fdd22bb7aa52 | 407 | q15_t * pScratch2) |
emilmont | 1:fdd22bb7aa52 | 408 | { |
emilmont | 1:fdd22bb7aa52 | 409 | |
emilmont | 1:fdd22bb7aa52 | 410 | q15_t *pOut = pDst; /* output pointer */ |
emilmont | 1:fdd22bb7aa52 | 411 | q15_t *pScr1 = pScratch1; /* Temporary pointer for scratch1 */ |
emilmont | 1:fdd22bb7aa52 | 412 | q15_t *pScr2 = pScratch2; /* Temporary pointer for scratch1 */ |
emilmont | 1:fdd22bb7aa52 | 413 | q31_t acc0, acc1, acc2, acc3; /* Accumulator */ |
emilmont | 1:fdd22bb7aa52 | 414 | q15_t *pIn1; /* inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 415 | q15_t *pIn2; /* inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 416 | q15_t *px; /* Intermediate inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 417 | q15_t *py; /* Intermediate inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 418 | uint32_t j, k, blkCnt; /* loop counter */ |
emilmont | 1:fdd22bb7aa52 | 419 | arm_status status; /* Status variable */ |
emilmont | 1:fdd22bb7aa52 | 420 | uint32_t tapCnt; /* loop count */ |
emilmont | 1:fdd22bb7aa52 | 421 | q15_t x10, x11, x20, x21; /* Temporary variables to hold srcA buffer */ |
emilmont | 1:fdd22bb7aa52 | 422 | q15_t y10, y11; /* Temporary variables to hold srcB buffer */ |
emilmont | 1:fdd22bb7aa52 | 423 | |
emilmont | 1:fdd22bb7aa52 | 424 | |
emilmont | 1:fdd22bb7aa52 | 425 | /* Check for range of output samples to be calculated */ |
emilmont | 1:fdd22bb7aa52 | 426 | if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u)))) |
emilmont | 1:fdd22bb7aa52 | 427 | { |
emilmont | 1:fdd22bb7aa52 | 428 | /* Set status as ARM_MATH_ARGUMENT_ERROR */ |
emilmont | 1:fdd22bb7aa52 | 429 | status = ARM_MATH_ARGUMENT_ERROR; |
emilmont | 1:fdd22bb7aa52 | 430 | } |
emilmont | 1:fdd22bb7aa52 | 431 | else |
emilmont | 1:fdd22bb7aa52 | 432 | { |
emilmont | 1:fdd22bb7aa52 | 433 | |
emilmont | 1:fdd22bb7aa52 | 434 | /* The algorithm implementation is based on the lengths of the inputs. */ |
emilmont | 1:fdd22bb7aa52 | 435 | /* srcB is always made to slide across srcA. */ |
emilmont | 1:fdd22bb7aa52 | 436 | /* So srcBLen is always considered as shorter or equal to srcALen */ |
emilmont | 1:fdd22bb7aa52 | 437 | if(srcALen >= srcBLen) |
emilmont | 1:fdd22bb7aa52 | 438 | { |
emilmont | 1:fdd22bb7aa52 | 439 | /* Initialization of inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 440 | pIn1 = pSrcA; |
emilmont | 1:fdd22bb7aa52 | 441 | |
emilmont | 1:fdd22bb7aa52 | 442 | /* Initialization of inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 443 | pIn2 = pSrcB; |
emilmont | 1:fdd22bb7aa52 | 444 | } |
emilmont | 1:fdd22bb7aa52 | 445 | else |
emilmont | 1:fdd22bb7aa52 | 446 | { |
emilmont | 1:fdd22bb7aa52 | 447 | /* Initialization of inputA pointer */ |
emilmont | 1:fdd22bb7aa52 | 448 | pIn1 = pSrcB; |
emilmont | 1:fdd22bb7aa52 | 449 | |
emilmont | 1:fdd22bb7aa52 | 450 | /* Initialization of inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 451 | pIn2 = pSrcA; |
emilmont | 1:fdd22bb7aa52 | 452 | |
emilmont | 1:fdd22bb7aa52 | 453 | /* srcBLen is always considered as shorter or equal to srcALen */ |
emilmont | 1:fdd22bb7aa52 | 454 | j = srcBLen; |
emilmont | 1:fdd22bb7aa52 | 455 | srcBLen = srcALen; |
emilmont | 1:fdd22bb7aa52 | 456 | srcALen = j; |
emilmont | 1:fdd22bb7aa52 | 457 | } |
emilmont | 1:fdd22bb7aa52 | 458 | |
emilmont | 1:fdd22bb7aa52 | 459 | /* Temporary pointer for scratch2 */ |
emilmont | 1:fdd22bb7aa52 | 460 | py = pScratch2; |
emilmont | 1:fdd22bb7aa52 | 461 | |
emilmont | 1:fdd22bb7aa52 | 462 | /* pointer to take end of scratch2 buffer */ |
emilmont | 1:fdd22bb7aa52 | 463 | pScr2 = pScratch2 + srcBLen - 1; |
emilmont | 1:fdd22bb7aa52 | 464 | |
emilmont | 1:fdd22bb7aa52 | 465 | /* points to smaller length sequence */ |
emilmont | 1:fdd22bb7aa52 | 466 | px = pIn2; |
emilmont | 1:fdd22bb7aa52 | 467 | |
emilmont | 1:fdd22bb7aa52 | 468 | /* Apply loop unrolling and do 4 Copies simultaneously. */ |
emilmont | 1:fdd22bb7aa52 | 469 | k = srcBLen >> 2u; |
emilmont | 1:fdd22bb7aa52 | 470 | |
emilmont | 1:fdd22bb7aa52 | 471 | /* First part of the processing with loop unrolling copies 4 data points at a time. |
emilmont | 1:fdd22bb7aa52 | 472 | ** a second loop below copies for the remaining 1 to 3 samples. */ |
emilmont | 1:fdd22bb7aa52 | 473 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 474 | { |
emilmont | 1:fdd22bb7aa52 | 475 | /* copy second buffer in reversal manner */ |
emilmont | 1:fdd22bb7aa52 | 476 | *pScr2-- = *px++; |
emilmont | 1:fdd22bb7aa52 | 477 | *pScr2-- = *px++; |
emilmont | 1:fdd22bb7aa52 | 478 | *pScr2-- = *px++; |
emilmont | 1:fdd22bb7aa52 | 479 | *pScr2-- = *px++; |
emilmont | 1:fdd22bb7aa52 | 480 | |
emilmont | 1:fdd22bb7aa52 | 481 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 482 | k--; |
emilmont | 1:fdd22bb7aa52 | 483 | } |
emilmont | 1:fdd22bb7aa52 | 484 | |
emilmont | 1:fdd22bb7aa52 | 485 | /* If the count is not a multiple of 4, copy remaining samples here. |
emilmont | 1:fdd22bb7aa52 | 486 | ** No loop unrolling is used. */ |
emilmont | 1:fdd22bb7aa52 | 487 | k = srcBLen % 0x4u; |
emilmont | 1:fdd22bb7aa52 | 488 | |
emilmont | 1:fdd22bb7aa52 | 489 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 490 | { |
emilmont | 1:fdd22bb7aa52 | 491 | /* copy second buffer in reversal manner for remaining samples */ |
emilmont | 1:fdd22bb7aa52 | 492 | *pScr2-- = *px++; |
emilmont | 1:fdd22bb7aa52 | 493 | |
emilmont | 1:fdd22bb7aa52 | 494 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 495 | k--; |
emilmont | 1:fdd22bb7aa52 | 496 | } |
emilmont | 1:fdd22bb7aa52 | 497 | |
emilmont | 1:fdd22bb7aa52 | 498 | /* Initialze temporary scratch pointer */ |
emilmont | 1:fdd22bb7aa52 | 499 | pScr1 = pScratch1; |
emilmont | 1:fdd22bb7aa52 | 500 | |
emilmont | 1:fdd22bb7aa52 | 501 | /* Fill (srcBLen - 1u) zeros in scratch buffer */ |
emilmont | 1:fdd22bb7aa52 | 502 | arm_fill_q15(0, pScr1, (srcBLen - 1u)); |
emilmont | 1:fdd22bb7aa52 | 503 | |
emilmont | 1:fdd22bb7aa52 | 504 | /* Update temporary scratch pointer */ |
emilmont | 1:fdd22bb7aa52 | 505 | pScr1 += (srcBLen - 1u); |
emilmont | 1:fdd22bb7aa52 | 506 | |
emilmont | 1:fdd22bb7aa52 | 507 | /* Copy bigger length sequence(srcALen) samples in scratch1 buffer */ |
emilmont | 1:fdd22bb7aa52 | 508 | |
emilmont | 1:fdd22bb7aa52 | 509 | |
emilmont | 1:fdd22bb7aa52 | 510 | /* Apply loop unrolling and do 4 Copies simultaneously. */ |
emilmont | 1:fdd22bb7aa52 | 511 | k = srcALen >> 2u; |
emilmont | 1:fdd22bb7aa52 | 512 | |
emilmont | 1:fdd22bb7aa52 | 513 | /* First part of the processing with loop unrolling copies 4 data points at a time. |
emilmont | 1:fdd22bb7aa52 | 514 | ** a second loop below copies for the remaining 1 to 3 samples. */ |
emilmont | 1:fdd22bb7aa52 | 515 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 516 | { |
emilmont | 1:fdd22bb7aa52 | 517 | /* copy second buffer in reversal manner */ |
emilmont | 1:fdd22bb7aa52 | 518 | *pScr1++ = *pIn1++; |
emilmont | 1:fdd22bb7aa52 | 519 | *pScr1++ = *pIn1++; |
emilmont | 1:fdd22bb7aa52 | 520 | *pScr1++ = *pIn1++; |
emilmont | 1:fdd22bb7aa52 | 521 | *pScr1++ = *pIn1++; |
emilmont | 1:fdd22bb7aa52 | 522 | |
emilmont | 1:fdd22bb7aa52 | 523 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 524 | k--; |
emilmont | 1:fdd22bb7aa52 | 525 | } |
emilmont | 1:fdd22bb7aa52 | 526 | |
emilmont | 1:fdd22bb7aa52 | 527 | /* If the count is not a multiple of 4, copy remaining samples here. |
emilmont | 1:fdd22bb7aa52 | 528 | ** No loop unrolling is used. */ |
emilmont | 1:fdd22bb7aa52 | 529 | k = srcALen % 0x4u; |
emilmont | 1:fdd22bb7aa52 | 530 | |
emilmont | 1:fdd22bb7aa52 | 531 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 532 | { |
emilmont | 1:fdd22bb7aa52 | 533 | /* copy second buffer in reversal manner for remaining samples */ |
emilmont | 1:fdd22bb7aa52 | 534 | *pScr1++ = *pIn1++; |
emilmont | 1:fdd22bb7aa52 | 535 | |
emilmont | 1:fdd22bb7aa52 | 536 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 537 | k--; |
emilmont | 1:fdd22bb7aa52 | 538 | } |
emilmont | 1:fdd22bb7aa52 | 539 | |
emilmont | 1:fdd22bb7aa52 | 540 | |
emilmont | 1:fdd22bb7aa52 | 541 | /* Apply loop unrolling and do 4 Copies simultaneously. */ |
emilmont | 1:fdd22bb7aa52 | 542 | k = (srcBLen - 1u) >> 2u; |
emilmont | 1:fdd22bb7aa52 | 543 | |
emilmont | 1:fdd22bb7aa52 | 544 | /* First part of the processing with loop unrolling copies 4 data points at a time. |
emilmont | 1:fdd22bb7aa52 | 545 | ** a second loop below copies for the remaining 1 to 3 samples. */ |
emilmont | 1:fdd22bb7aa52 | 546 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 547 | { |
emilmont | 1:fdd22bb7aa52 | 548 | /* copy second buffer in reversal manner */ |
emilmont | 1:fdd22bb7aa52 | 549 | *pScr1++ = 0; |
emilmont | 1:fdd22bb7aa52 | 550 | *pScr1++ = 0; |
emilmont | 1:fdd22bb7aa52 | 551 | *pScr1++ = 0; |
emilmont | 1:fdd22bb7aa52 | 552 | *pScr1++ = 0; |
emilmont | 1:fdd22bb7aa52 | 553 | |
emilmont | 1:fdd22bb7aa52 | 554 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 555 | k--; |
emilmont | 1:fdd22bb7aa52 | 556 | } |
emilmont | 1:fdd22bb7aa52 | 557 | |
emilmont | 1:fdd22bb7aa52 | 558 | /* If the count is not a multiple of 4, copy remaining samples here. |
emilmont | 1:fdd22bb7aa52 | 559 | ** No loop unrolling is used. */ |
emilmont | 1:fdd22bb7aa52 | 560 | k = (srcBLen - 1u) % 0x4u; |
emilmont | 1:fdd22bb7aa52 | 561 | |
emilmont | 1:fdd22bb7aa52 | 562 | while(k > 0u) |
emilmont | 1:fdd22bb7aa52 | 563 | { |
emilmont | 1:fdd22bb7aa52 | 564 | /* copy second buffer in reversal manner for remaining samples */ |
emilmont | 1:fdd22bb7aa52 | 565 | *pScr1++ = 0; |
emilmont | 1:fdd22bb7aa52 | 566 | |
emilmont | 1:fdd22bb7aa52 | 567 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 568 | k--; |
emilmont | 1:fdd22bb7aa52 | 569 | } |
emilmont | 1:fdd22bb7aa52 | 570 | |
emilmont | 1:fdd22bb7aa52 | 571 | |
emilmont | 1:fdd22bb7aa52 | 572 | /* Initialization of pIn2 pointer */ |
emilmont | 1:fdd22bb7aa52 | 573 | pIn2 = py; |
emilmont | 1:fdd22bb7aa52 | 574 | |
emilmont | 1:fdd22bb7aa52 | 575 | pScratch1 += firstIndex; |
emilmont | 1:fdd22bb7aa52 | 576 | |
emilmont | 1:fdd22bb7aa52 | 577 | pOut = pDst + firstIndex; |
emilmont | 1:fdd22bb7aa52 | 578 | |
emilmont | 1:fdd22bb7aa52 | 579 | /* Actual convolution process starts here */ |
emilmont | 1:fdd22bb7aa52 | 580 | blkCnt = (numPoints) >> 2; |
emilmont | 1:fdd22bb7aa52 | 581 | |
emilmont | 1:fdd22bb7aa52 | 582 | while(blkCnt > 0) |
emilmont | 1:fdd22bb7aa52 | 583 | { |
emilmont | 1:fdd22bb7aa52 | 584 | /* Initialze temporary scratch pointer as scratch1 */ |
emilmont | 1:fdd22bb7aa52 | 585 | pScr1 = pScratch1; |
emilmont | 1:fdd22bb7aa52 | 586 | |
emilmont | 1:fdd22bb7aa52 | 587 | /* Clear Accumlators */ |
emilmont | 1:fdd22bb7aa52 | 588 | acc0 = 0; |
emilmont | 1:fdd22bb7aa52 | 589 | acc1 = 0; |
emilmont | 1:fdd22bb7aa52 | 590 | acc2 = 0; |
emilmont | 1:fdd22bb7aa52 | 591 | acc3 = 0; |
emilmont | 1:fdd22bb7aa52 | 592 | |
emilmont | 1:fdd22bb7aa52 | 593 | /* Read two samples from scratch1 buffer */ |
emilmont | 1:fdd22bb7aa52 | 594 | x10 = *pScr1++; |
emilmont | 1:fdd22bb7aa52 | 595 | x11 = *pScr1++; |
emilmont | 1:fdd22bb7aa52 | 596 | |
emilmont | 1:fdd22bb7aa52 | 597 | /* Read next two samples from scratch1 buffer */ |
emilmont | 1:fdd22bb7aa52 | 598 | x20 = *pScr1++; |
emilmont | 1:fdd22bb7aa52 | 599 | x21 = *pScr1++; |
emilmont | 1:fdd22bb7aa52 | 600 | |
emilmont | 1:fdd22bb7aa52 | 601 | tapCnt = (srcBLen) >> 2u; |
emilmont | 1:fdd22bb7aa52 | 602 | |
emilmont | 1:fdd22bb7aa52 | 603 | while(tapCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 604 | { |
emilmont | 1:fdd22bb7aa52 | 605 | |
emilmont | 1:fdd22bb7aa52 | 606 | /* Read two samples from smaller buffer */ |
emilmont | 1:fdd22bb7aa52 | 607 | y10 = *pIn2; |
emilmont | 1:fdd22bb7aa52 | 608 | y11 = *(pIn2 + 1u); |
emilmont | 1:fdd22bb7aa52 | 609 | |
emilmont | 1:fdd22bb7aa52 | 610 | /* multiply and accumlate */ |
emilmont | 1:fdd22bb7aa52 | 611 | acc0 += (q31_t) x10 *y10; |
emilmont | 1:fdd22bb7aa52 | 612 | acc0 += (q31_t) x11 *y11; |
emilmont | 1:fdd22bb7aa52 | 613 | acc2 += (q31_t) x20 *y10; |
emilmont | 1:fdd22bb7aa52 | 614 | acc2 += (q31_t) x21 *y11; |
emilmont | 1:fdd22bb7aa52 | 615 | |
emilmont | 1:fdd22bb7aa52 | 616 | /* multiply and accumlate */ |
emilmont | 1:fdd22bb7aa52 | 617 | acc1 += (q31_t) x11 *y10; |
emilmont | 1:fdd22bb7aa52 | 618 | acc1 += (q31_t) x20 *y11; |
emilmont | 1:fdd22bb7aa52 | 619 | |
emilmont | 1:fdd22bb7aa52 | 620 | /* Read next two samples from scratch1 buffer */ |
emilmont | 1:fdd22bb7aa52 | 621 | x10 = *pScr1; |
emilmont | 1:fdd22bb7aa52 | 622 | x11 = *(pScr1 + 1u); |
emilmont | 1:fdd22bb7aa52 | 623 | |
emilmont | 1:fdd22bb7aa52 | 624 | /* multiply and accumlate */ |
emilmont | 1:fdd22bb7aa52 | 625 | acc3 += (q31_t) x21 *y10; |
emilmont | 1:fdd22bb7aa52 | 626 | acc3 += (q31_t) x10 *y11; |
emilmont | 1:fdd22bb7aa52 | 627 | |
emilmont | 1:fdd22bb7aa52 | 628 | /* Read next two samples from scratch2 buffer */ |
emilmont | 1:fdd22bb7aa52 | 629 | y10 = *(pIn2 + 2u); |
emilmont | 1:fdd22bb7aa52 | 630 | y11 = *(pIn2 + 3u); |
emilmont | 1:fdd22bb7aa52 | 631 | |
emilmont | 1:fdd22bb7aa52 | 632 | /* multiply and accumlate */ |
emilmont | 1:fdd22bb7aa52 | 633 | acc0 += (q31_t) x20 *y10; |
emilmont | 1:fdd22bb7aa52 | 634 | acc0 += (q31_t) x21 *y11; |
emilmont | 1:fdd22bb7aa52 | 635 | acc2 += (q31_t) x10 *y10; |
emilmont | 1:fdd22bb7aa52 | 636 | acc2 += (q31_t) x11 *y11; |
emilmont | 1:fdd22bb7aa52 | 637 | acc1 += (q31_t) x21 *y10; |
emilmont | 1:fdd22bb7aa52 | 638 | acc1 += (q31_t) x10 *y11; |
emilmont | 1:fdd22bb7aa52 | 639 | |
emilmont | 1:fdd22bb7aa52 | 640 | /* Read next two samples from scratch1 buffer */ |
emilmont | 1:fdd22bb7aa52 | 641 | x20 = *(pScr1 + 2); |
emilmont | 1:fdd22bb7aa52 | 642 | x21 = *(pScr1 + 3); |
emilmont | 1:fdd22bb7aa52 | 643 | |
emilmont | 1:fdd22bb7aa52 | 644 | /* multiply and accumlate */ |
emilmont | 1:fdd22bb7aa52 | 645 | acc3 += (q31_t) x11 *y10; |
emilmont | 1:fdd22bb7aa52 | 646 | acc3 += (q31_t) x20 *y11; |
emilmont | 1:fdd22bb7aa52 | 647 | |
emilmont | 1:fdd22bb7aa52 | 648 | /* update scratch pointers */ |
emilmont | 1:fdd22bb7aa52 | 649 | pIn2 += 4u; |
emilmont | 1:fdd22bb7aa52 | 650 | pScr1 += 4u; |
emilmont | 1:fdd22bb7aa52 | 651 | |
emilmont | 1:fdd22bb7aa52 | 652 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 653 | tapCnt--; |
emilmont | 1:fdd22bb7aa52 | 654 | } |
emilmont | 1:fdd22bb7aa52 | 655 | |
emilmont | 1:fdd22bb7aa52 | 656 | /* Update scratch pointer for remaining samples of smaller length sequence */ |
emilmont | 1:fdd22bb7aa52 | 657 | pScr1 -= 4u; |
emilmont | 1:fdd22bb7aa52 | 658 | |
emilmont | 1:fdd22bb7aa52 | 659 | /* apply same above for remaining samples of smaller length sequence */ |
emilmont | 1:fdd22bb7aa52 | 660 | tapCnt = (srcBLen) & 3u; |
emilmont | 1:fdd22bb7aa52 | 661 | |
emilmont | 1:fdd22bb7aa52 | 662 | while(tapCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 663 | { |
emilmont | 1:fdd22bb7aa52 | 664 | /* accumlate the results */ |
emilmont | 1:fdd22bb7aa52 | 665 | acc0 += (*pScr1++ * *pIn2); |
emilmont | 1:fdd22bb7aa52 | 666 | acc1 += (*pScr1++ * *pIn2); |
emilmont | 1:fdd22bb7aa52 | 667 | acc2 += (*pScr1++ * *pIn2); |
emilmont | 1:fdd22bb7aa52 | 668 | acc3 += (*pScr1++ * *pIn2++); |
emilmont | 1:fdd22bb7aa52 | 669 | |
emilmont | 1:fdd22bb7aa52 | 670 | pScr1 -= 3u; |
emilmont | 1:fdd22bb7aa52 | 671 | |
emilmont | 1:fdd22bb7aa52 | 672 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 673 | tapCnt--; |
emilmont | 1:fdd22bb7aa52 | 674 | } |
emilmont | 1:fdd22bb7aa52 | 675 | |
emilmont | 1:fdd22bb7aa52 | 676 | blkCnt--; |
emilmont | 1:fdd22bb7aa52 | 677 | |
emilmont | 1:fdd22bb7aa52 | 678 | |
emilmont | 1:fdd22bb7aa52 | 679 | /* Store the results in the accumulators in the destination buffer. */ |
emilmont | 1:fdd22bb7aa52 | 680 | *pOut++ = __SSAT((acc0 >> 15), 16); |
emilmont | 1:fdd22bb7aa52 | 681 | *pOut++ = __SSAT((acc1 >> 15), 16); |
emilmont | 1:fdd22bb7aa52 | 682 | *pOut++ = __SSAT((acc2 >> 15), 16); |
emilmont | 1:fdd22bb7aa52 | 683 | *pOut++ = __SSAT((acc3 >> 15), 16); |
emilmont | 1:fdd22bb7aa52 | 684 | |
emilmont | 1:fdd22bb7aa52 | 685 | /* Initialization of inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 686 | pIn2 = py; |
emilmont | 1:fdd22bb7aa52 | 687 | |
emilmont | 1:fdd22bb7aa52 | 688 | pScratch1 += 4u; |
emilmont | 1:fdd22bb7aa52 | 689 | |
emilmont | 1:fdd22bb7aa52 | 690 | } |
emilmont | 1:fdd22bb7aa52 | 691 | |
emilmont | 1:fdd22bb7aa52 | 692 | |
emilmont | 1:fdd22bb7aa52 | 693 | blkCnt = numPoints & 0x3; |
emilmont | 1:fdd22bb7aa52 | 694 | |
emilmont | 1:fdd22bb7aa52 | 695 | /* Calculate convolution for remaining samples of Bigger length sequence */ |
emilmont | 1:fdd22bb7aa52 | 696 | while(blkCnt > 0) |
emilmont | 1:fdd22bb7aa52 | 697 | { |
emilmont | 1:fdd22bb7aa52 | 698 | /* Initialze temporary scratch pointer as scratch1 */ |
emilmont | 1:fdd22bb7aa52 | 699 | pScr1 = pScratch1; |
emilmont | 1:fdd22bb7aa52 | 700 | |
emilmont | 1:fdd22bb7aa52 | 701 | /* Clear Accumlators */ |
emilmont | 1:fdd22bb7aa52 | 702 | acc0 = 0; |
emilmont | 1:fdd22bb7aa52 | 703 | |
emilmont | 1:fdd22bb7aa52 | 704 | tapCnt = (srcBLen) >> 1u; |
emilmont | 1:fdd22bb7aa52 | 705 | |
emilmont | 1:fdd22bb7aa52 | 706 | while(tapCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 707 | { |
emilmont | 1:fdd22bb7aa52 | 708 | |
emilmont | 1:fdd22bb7aa52 | 709 | /* Read next two samples from scratch1 buffer */ |
emilmont | 1:fdd22bb7aa52 | 710 | x10 = *pScr1++; |
emilmont | 1:fdd22bb7aa52 | 711 | x11 = *pScr1++; |
emilmont | 1:fdd22bb7aa52 | 712 | |
emilmont | 1:fdd22bb7aa52 | 713 | /* Read two samples from smaller buffer */ |
emilmont | 1:fdd22bb7aa52 | 714 | y10 = *pIn2++; |
emilmont | 1:fdd22bb7aa52 | 715 | y11 = *pIn2++; |
emilmont | 1:fdd22bb7aa52 | 716 | |
emilmont | 1:fdd22bb7aa52 | 717 | /* multiply and accumlate */ |
emilmont | 1:fdd22bb7aa52 | 718 | acc0 += (q31_t) x10 *y10; |
emilmont | 1:fdd22bb7aa52 | 719 | acc0 += (q31_t) x11 *y11; |
emilmont | 1:fdd22bb7aa52 | 720 | |
emilmont | 1:fdd22bb7aa52 | 721 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 722 | tapCnt--; |
emilmont | 1:fdd22bb7aa52 | 723 | } |
emilmont | 1:fdd22bb7aa52 | 724 | |
emilmont | 1:fdd22bb7aa52 | 725 | tapCnt = (srcBLen) & 1u; |
emilmont | 1:fdd22bb7aa52 | 726 | |
emilmont | 1:fdd22bb7aa52 | 727 | /* apply same above for remaining samples of smaller length sequence */ |
emilmont | 1:fdd22bb7aa52 | 728 | while(tapCnt > 0u) |
emilmont | 1:fdd22bb7aa52 | 729 | { |
emilmont | 1:fdd22bb7aa52 | 730 | |
emilmont | 1:fdd22bb7aa52 | 731 | /* accumlate the results */ |
emilmont | 1:fdd22bb7aa52 | 732 | acc0 += (*pScr1++ * *pIn2++); |
emilmont | 1:fdd22bb7aa52 | 733 | |
emilmont | 1:fdd22bb7aa52 | 734 | /* Decrement the loop counter */ |
emilmont | 1:fdd22bb7aa52 | 735 | tapCnt--; |
emilmont | 1:fdd22bb7aa52 | 736 | } |
emilmont | 1:fdd22bb7aa52 | 737 | |
emilmont | 1:fdd22bb7aa52 | 738 | blkCnt--; |
emilmont | 1:fdd22bb7aa52 | 739 | |
emilmont | 1:fdd22bb7aa52 | 740 | /* Store the result in the accumulator in the destination buffer. */ |
emilmont | 1:fdd22bb7aa52 | 741 | *pOut++ = (q15_t) (__SSAT((acc0 >> 15), 16)); |
emilmont | 1:fdd22bb7aa52 | 742 | |
emilmont | 1:fdd22bb7aa52 | 743 | /* Initialization of inputB pointer */ |
emilmont | 1:fdd22bb7aa52 | 744 | pIn2 = py; |
emilmont | 1:fdd22bb7aa52 | 745 | |
emilmont | 1:fdd22bb7aa52 | 746 | pScratch1 += 1u; |
emilmont | 1:fdd22bb7aa52 | 747 | |
emilmont | 1:fdd22bb7aa52 | 748 | } |
emilmont | 1:fdd22bb7aa52 | 749 | |
emilmont | 1:fdd22bb7aa52 | 750 | /* set status as ARM_MATH_SUCCESS */ |
emilmont | 1:fdd22bb7aa52 | 751 | status = ARM_MATH_SUCCESS; |
emilmont | 1:fdd22bb7aa52 | 752 | |
emilmont | 1:fdd22bb7aa52 | 753 | } |
emilmont | 1:fdd22bb7aa52 | 754 | |
emilmont | 1:fdd22bb7aa52 | 755 | /* Return to application */ |
emilmont | 1:fdd22bb7aa52 | 756 | return (status); |
emilmont | 1:fdd22bb7aa52 | 757 | } |
emilmont | 1:fdd22bb7aa52 | 758 | |
emilmont | 1:fdd22bb7aa52 | 759 | #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */ |
emilmont | 1:fdd22bb7aa52 | 760 | |
emilmont | 1:fdd22bb7aa52 | 761 | /** |
emilmont | 1:fdd22bb7aa52 | 762 | * @} end of PartialConv group |
emilmont | 1:fdd22bb7aa52 | 763 | */ |