mbed official / mbed-dsp

CMSIS DSP library

Dependents:   KL25Z_FFT_Demo Hat_Board_v5_1 KL25Z_FFT_Demo_tony KL25Z_FFT_Demo_tony ... more

Fork of mbed-dsp by mbed official

cmsis_dsp/FilteringFunctions/arm_conv_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?

UserRevisionLine numberNew 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_opt_q15.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 1:fdd22bb7aa52 10 * Description: Convolution of Q15 sequences.
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 1.0.11 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 * Version 0.0.7 2010/06/10
emilmont 1:fdd22bb7aa52 36 * Misra-C changes done
emilmont 1:fdd22bb7aa52 37 *
emilmont 1:fdd22bb7aa52 38 * -------------------------------------------------------------------- */
emilmont 1:fdd22bb7aa52 39
emilmont 1:fdd22bb7aa52 40 #include "arm_math.h"
emilmont 1:fdd22bb7aa52 41
emilmont 1:fdd22bb7aa52 42 /**
emilmont 1:fdd22bb7aa52 43 * @ingroup groupFilters
emilmont 1:fdd22bb7aa52 44 */
emilmont 1:fdd22bb7aa52 45
emilmont 1:fdd22bb7aa52 46 /**
emilmont 1:fdd22bb7aa52 47 * @addtogroup Conv
emilmont 1:fdd22bb7aa52 48 * @{
emilmont 1:fdd22bb7aa52 49 */
emilmont 1:fdd22bb7aa52 50
emilmont 1:fdd22bb7aa52 51 /**
emilmont 1:fdd22bb7aa52 52 * @brief Convolution of Q15 sequences.
emilmont 1:fdd22bb7aa52 53 * @param[in] *pSrcA points to the first input sequence.
emilmont 1:fdd22bb7aa52 54 * @param[in] srcALen length of the first input sequence.
emilmont 1:fdd22bb7aa52 55 * @param[in] *pSrcB points to the second input sequence.
emilmont 1:fdd22bb7aa52 56 * @param[in] srcBLen length of the second input sequence.
emilmont 1:fdd22bb7aa52 57 * @param[out] *pDst points to the location where the output result is written. Length srcALen+srcBLen-1.
emilmont 1:fdd22bb7aa52 58 * @param[in] *pScratch1 points to scratch buffer of size max(srcALen, srcBLen) + 2*min(srcALen, srcBLen) - 2.
emilmont 1:fdd22bb7aa52 59 * @param[in] *pScratch2 points to scratch buffer of size min(srcALen, srcBLen).
emilmont 1:fdd22bb7aa52 60 * @return none.
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 * @details
emilmont 1:fdd22bb7aa52 68 * <b>Scaling and Overflow Behavior:</b>
emilmont 1:fdd22bb7aa52 69 *
emilmont 1:fdd22bb7aa52 70 * \par
emilmont 1:fdd22bb7aa52 71 * The function is implemented using a 64-bit internal accumulator.
emilmont 1:fdd22bb7aa52 72 * Both inputs are in 1.15 format and multiplications yield a 2.30 result.
emilmont 1:fdd22bb7aa52 73 * The 2.30 intermediate results are accumulated in a 64-bit accumulator in 34.30 format.
emilmont 1:fdd22bb7aa52 74 * This approach provides 33 guard bits and there is no risk of overflow.
emilmont 1:fdd22bb7aa52 75 * The 34.30 result is then truncated to 34.15 format by discarding the low 15 bits and then saturated to 1.15 format.
emilmont 1:fdd22bb7aa52 76 *
emilmont 1:fdd22bb7aa52 77 *
emilmont 1:fdd22bb7aa52 78 * \par
emilmont 1:fdd22bb7aa52 79 * Refer to <code>arm_conv_fast_q15()</code> for a faster but less precise version of this function for Cortex-M3 and Cortex-M4.
emilmont 1:fdd22bb7aa52 80 *
emilmont 1:fdd22bb7aa52 81 *
emilmont 1:fdd22bb7aa52 82 */
emilmont 1:fdd22bb7aa52 83
emilmont 1:fdd22bb7aa52 84 void arm_conv_opt_q15(
emilmont 1:fdd22bb7aa52 85 q15_t * pSrcA,
emilmont 1:fdd22bb7aa52 86 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 87 q15_t * pSrcB,
emilmont 1:fdd22bb7aa52 88 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 89 q15_t * pDst,
emilmont 1:fdd22bb7aa52 90 q15_t * pScratch1,
emilmont 1:fdd22bb7aa52 91 q15_t * pScratch2)
emilmont 1:fdd22bb7aa52 92 {
emilmont 1:fdd22bb7aa52 93 q63_t acc0, acc1, acc2, acc3; /* Accumulator */
emilmont 1:fdd22bb7aa52 94 q31_t x1, x2, x3; /* Temporary variables to hold state and coefficient values */
emilmont 1:fdd22bb7aa52 95 q31_t y1, y2; /* State variables */
emilmont 1:fdd22bb7aa52 96 q15_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 97 q15_t *pScr1 = pScratch1; /* Temporary pointer for scratch1 */
emilmont 1:fdd22bb7aa52 98 q15_t *pScr2 = pScratch2; /* Temporary pointer for scratch1 */
emilmont 1:fdd22bb7aa52 99 q15_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 100 q15_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 101 q15_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 102 q15_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 103 uint32_t j, k, blkCnt; /* loop counter */
emilmont 1:fdd22bb7aa52 104 uint32_t tapCnt; /* loop count */
emilmont 1:fdd22bb7aa52 105 #ifdef UNALIGNED_SUPPORT_DISABLE
emilmont 1:fdd22bb7aa52 106
emilmont 1:fdd22bb7aa52 107 q15_t a, b;
emilmont 1:fdd22bb7aa52 108
emilmont 1:fdd22bb7aa52 109 #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */
emilmont 1:fdd22bb7aa52 110
emilmont 1:fdd22bb7aa52 111 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 112 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 113 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 114 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 115 {
emilmont 1:fdd22bb7aa52 116 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 117 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 118
emilmont 1:fdd22bb7aa52 119 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 120 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 121
emilmont 1:fdd22bb7aa52 122 }
emilmont 1:fdd22bb7aa52 123 else
emilmont 1:fdd22bb7aa52 124 {
emilmont 1:fdd22bb7aa52 125 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 126 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 127
emilmont 1:fdd22bb7aa52 128 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 129 pIn2 = pSrcA;
emilmont 1:fdd22bb7aa52 130
emilmont 1:fdd22bb7aa52 131 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 132 j = srcBLen;
emilmont 1:fdd22bb7aa52 133 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 134 srcALen = j;
emilmont 1:fdd22bb7aa52 135 }
emilmont 1:fdd22bb7aa52 136
emilmont 1:fdd22bb7aa52 137 /* pointer to take end of scratch2 buffer */
emilmont 1:fdd22bb7aa52 138 pScr2 = pScratch2 + srcBLen - 1;
emilmont 1:fdd22bb7aa52 139
emilmont 1:fdd22bb7aa52 140 /* points to smaller length sequence */
emilmont 1:fdd22bb7aa52 141 px = pIn2;
emilmont 1:fdd22bb7aa52 142
emilmont 1:fdd22bb7aa52 143 /* Apply loop unrolling and do 4 Copies simultaneously. */
emilmont 1:fdd22bb7aa52 144 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 145
emilmont 1:fdd22bb7aa52 146 /* First part of the processing with loop unrolling copies 4 data points at a time.
emilmont 1:fdd22bb7aa52 147 ** a second loop below copies for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 148 /* Copy smaller length input sequence in reverse order into second scratch buffer */
emilmont 1:fdd22bb7aa52 149 while(k > 0u)
emilmont 1:fdd22bb7aa52 150 {
emilmont 1:fdd22bb7aa52 151 /* copy second buffer in reversal manner */
emilmont 1:fdd22bb7aa52 152 *pScr2-- = *px++;
emilmont 1:fdd22bb7aa52 153 *pScr2-- = *px++;
emilmont 1:fdd22bb7aa52 154 *pScr2-- = *px++;
emilmont 1:fdd22bb7aa52 155 *pScr2-- = *px++;
emilmont 1:fdd22bb7aa52 156
emilmont 1:fdd22bb7aa52 157 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 158 k--;
emilmont 1:fdd22bb7aa52 159 }
emilmont 1:fdd22bb7aa52 160
emilmont 1:fdd22bb7aa52 161 /* If the count is not a multiple of 4, copy remaining samples here.
emilmont 1:fdd22bb7aa52 162 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 163 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 164
emilmont 1:fdd22bb7aa52 165 while(k > 0u)
emilmont 1:fdd22bb7aa52 166 {
emilmont 1:fdd22bb7aa52 167 /* copy second buffer in reversal manner for remaining samples */
emilmont 1:fdd22bb7aa52 168 *pScr2-- = *px++;
emilmont 1:fdd22bb7aa52 169
emilmont 1:fdd22bb7aa52 170 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 171 k--;
emilmont 1:fdd22bb7aa52 172 }
emilmont 1:fdd22bb7aa52 173
emilmont 1:fdd22bb7aa52 174 /* Initialze temporary scratch pointer */
emilmont 1:fdd22bb7aa52 175 pScr1 = pScratch1;
emilmont 1:fdd22bb7aa52 176
emilmont 1:fdd22bb7aa52 177 /* Assuming scratch1 buffer is aligned by 32-bit */
emilmont 1:fdd22bb7aa52 178 /* Fill (srcBLen - 1u) zeros in scratch buffer */
emilmont 1:fdd22bb7aa52 179 arm_fill_q15(0, pScr1, (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 180
emilmont 1:fdd22bb7aa52 181 /* Update temporary scratch pointer */
emilmont 1:fdd22bb7aa52 182 pScr1 += (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 183
emilmont 1:fdd22bb7aa52 184 /* Copy bigger length sequence(srcALen) samples in scratch1 buffer */
emilmont 1:fdd22bb7aa52 185
emilmont 1:fdd22bb7aa52 186 #ifndef UNALIGNED_SUPPORT_DISABLE
emilmont 1:fdd22bb7aa52 187
emilmont 1:fdd22bb7aa52 188 /* Copy (srcALen) samples in scratch buffer */
emilmont 1:fdd22bb7aa52 189 arm_copy_q15(pIn1, pScr1, srcALen);
emilmont 1:fdd22bb7aa52 190
emilmont 1:fdd22bb7aa52 191 /* Update pointers */
emilmont 1:fdd22bb7aa52 192 pScr1 += srcALen;
emilmont 1:fdd22bb7aa52 193
emilmont 1:fdd22bb7aa52 194 #else
emilmont 1:fdd22bb7aa52 195
emilmont 1:fdd22bb7aa52 196 /* Apply loop unrolling and do 4 Copies simultaneously. */
emilmont 1:fdd22bb7aa52 197 k = srcALen >> 2u;
emilmont 1:fdd22bb7aa52 198
emilmont 1:fdd22bb7aa52 199 /* First part of the processing with loop unrolling copies 4 data points at a time.
emilmont 1:fdd22bb7aa52 200 ** a second loop below copies for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 201 while(k > 0u)
emilmont 1:fdd22bb7aa52 202 {
emilmont 1:fdd22bb7aa52 203 /* copy second buffer in reversal manner */
emilmont 1:fdd22bb7aa52 204 *pScr1++ = *pIn1++;
emilmont 1:fdd22bb7aa52 205 *pScr1++ = *pIn1++;
emilmont 1:fdd22bb7aa52 206 *pScr1++ = *pIn1++;
emilmont 1:fdd22bb7aa52 207 *pScr1++ = *pIn1++;
emilmont 1:fdd22bb7aa52 208
emilmont 1:fdd22bb7aa52 209 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 210 k--;
emilmont 1:fdd22bb7aa52 211 }
emilmont 1:fdd22bb7aa52 212
emilmont 1:fdd22bb7aa52 213 /* If the count is not a multiple of 4, copy remaining samples here.
emilmont 1:fdd22bb7aa52 214 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 215 k = srcALen % 0x4u;
emilmont 1:fdd22bb7aa52 216
emilmont 1:fdd22bb7aa52 217 while(k > 0u)
emilmont 1:fdd22bb7aa52 218 {
emilmont 1:fdd22bb7aa52 219 /* copy second buffer in reversal manner for remaining samples */
emilmont 1:fdd22bb7aa52 220 *pScr1++ = *pIn1++;
emilmont 1:fdd22bb7aa52 221
emilmont 1:fdd22bb7aa52 222 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 223 k--;
emilmont 1:fdd22bb7aa52 224 }
emilmont 1:fdd22bb7aa52 225
emilmont 1:fdd22bb7aa52 226 #endif
emilmont 1:fdd22bb7aa52 227
emilmont 1:fdd22bb7aa52 228
emilmont 1:fdd22bb7aa52 229 #ifndef UNALIGNED_SUPPORT_DISABLE
emilmont 1:fdd22bb7aa52 230
emilmont 1:fdd22bb7aa52 231 /* Fill (srcBLen - 1u) zeros at end of scratch buffer */
emilmont 1:fdd22bb7aa52 232 arm_fill_q15(0, pScr1, (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 233
emilmont 1:fdd22bb7aa52 234 /* Update pointer */
emilmont 1:fdd22bb7aa52 235 pScr1 += (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 236
emilmont 1:fdd22bb7aa52 237 #else
emilmont 1:fdd22bb7aa52 238
emilmont 1:fdd22bb7aa52 239 /* Apply loop unrolling and do 4 Copies simultaneously. */
emilmont 1:fdd22bb7aa52 240 k = (srcBLen - 1u) >> 2u;
emilmont 1:fdd22bb7aa52 241
emilmont 1:fdd22bb7aa52 242 /* First part of the processing with loop unrolling copies 4 data points at a time.
emilmont 1:fdd22bb7aa52 243 ** a second loop below copies for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 244 while(k > 0u)
emilmont 1:fdd22bb7aa52 245 {
emilmont 1:fdd22bb7aa52 246 /* copy second buffer in reversal manner */
emilmont 1:fdd22bb7aa52 247 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 248 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 249 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 250 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 251
emilmont 1:fdd22bb7aa52 252 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 253 k--;
emilmont 1:fdd22bb7aa52 254 }
emilmont 1:fdd22bb7aa52 255
emilmont 1:fdd22bb7aa52 256 /* If the count is not a multiple of 4, copy remaining samples here.
emilmont 1:fdd22bb7aa52 257 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 258 k = (srcBLen - 1u) % 0x4u;
emilmont 1:fdd22bb7aa52 259
emilmont 1:fdd22bb7aa52 260 while(k > 0u)
emilmont 1:fdd22bb7aa52 261 {
emilmont 1:fdd22bb7aa52 262 /* copy second buffer in reversal manner for remaining samples */
emilmont 1:fdd22bb7aa52 263 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 264
emilmont 1:fdd22bb7aa52 265 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 266 k--;
emilmont 1:fdd22bb7aa52 267 }
emilmont 1:fdd22bb7aa52 268
emilmont 1:fdd22bb7aa52 269 #endif
emilmont 1:fdd22bb7aa52 270
emilmont 1:fdd22bb7aa52 271 /* Temporary pointer for scratch2 */
emilmont 1:fdd22bb7aa52 272 py = pScratch2;
emilmont 1:fdd22bb7aa52 273
emilmont 1:fdd22bb7aa52 274
emilmont 1:fdd22bb7aa52 275 /* Initialization of pIn2 pointer */
emilmont 1:fdd22bb7aa52 276 pIn2 = py;
emilmont 1:fdd22bb7aa52 277
emilmont 1:fdd22bb7aa52 278 /* First part of the processing with loop unrolling process 4 data points at a time.
emilmont 1:fdd22bb7aa52 279 ** a second loop below process for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 280
emilmont 1:fdd22bb7aa52 281 /* Actual convolution process starts here */
emilmont 1:fdd22bb7aa52 282 blkCnt = (srcALen + srcBLen - 1u) >> 2;
emilmont 1:fdd22bb7aa52 283
emilmont 1:fdd22bb7aa52 284 while(blkCnt > 0)
emilmont 1:fdd22bb7aa52 285 {
emilmont 1:fdd22bb7aa52 286 /* Initialze temporary scratch pointer as scratch1 */
emilmont 1:fdd22bb7aa52 287 pScr1 = pScratch1;
emilmont 1:fdd22bb7aa52 288
emilmont 1:fdd22bb7aa52 289 /* Clear Accumlators */
emilmont 1:fdd22bb7aa52 290 acc0 = 0;
emilmont 1:fdd22bb7aa52 291 acc1 = 0;
emilmont 1:fdd22bb7aa52 292 acc2 = 0;
emilmont 1:fdd22bb7aa52 293 acc3 = 0;
emilmont 1:fdd22bb7aa52 294
emilmont 1:fdd22bb7aa52 295 /* Read two samples from scratch1 buffer */
emilmont 1:fdd22bb7aa52 296 x1 = *__SIMD32(pScr1)++;
emilmont 1:fdd22bb7aa52 297
emilmont 1:fdd22bb7aa52 298 /* Read next two samples from scratch1 buffer */
emilmont 1:fdd22bb7aa52 299 x2 = *__SIMD32(pScr1)++;
emilmont 1:fdd22bb7aa52 300
emilmont 1:fdd22bb7aa52 301 tapCnt = (srcBLen) >> 2u;
emilmont 1:fdd22bb7aa52 302
emilmont 1:fdd22bb7aa52 303 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 304 {
emilmont 1:fdd22bb7aa52 305
emilmont 1:fdd22bb7aa52 306 #ifndef UNALIGNED_SUPPORT_DISABLE
emilmont 1:fdd22bb7aa52 307
emilmont 1:fdd22bb7aa52 308 /* Read four samples from smaller buffer */
emilmont 1:fdd22bb7aa52 309 y1 = _SIMD32_OFFSET(pIn2);
emilmont 1:fdd22bb7aa52 310 y2 = _SIMD32_OFFSET(pIn2 + 2u);
emilmont 1:fdd22bb7aa52 311
emilmont 1:fdd22bb7aa52 312 /* multiply and accumlate */
emilmont 1:fdd22bb7aa52 313 acc0 = __SMLALD(x1, y1, acc0);
emilmont 1:fdd22bb7aa52 314 acc2 = __SMLALD(x2, y1, acc2);
emilmont 1:fdd22bb7aa52 315
emilmont 1:fdd22bb7aa52 316 /* pack input data */
emilmont 1:fdd22bb7aa52 317 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 318 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 319 #else
emilmont 1:fdd22bb7aa52 320 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 321 #endif
emilmont 1:fdd22bb7aa52 322
emilmont 1:fdd22bb7aa52 323 /* multiply and accumlate */
emilmont 1:fdd22bb7aa52 324 acc1 = __SMLALDX(x3, y1, acc1);
emilmont 1:fdd22bb7aa52 325
emilmont 1:fdd22bb7aa52 326 /* Read next two samples from scratch1 buffer */
emilmont 1:fdd22bb7aa52 327 x1 = _SIMD32_OFFSET(pScr1);
emilmont 1:fdd22bb7aa52 328
emilmont 1:fdd22bb7aa52 329 /* multiply and accumlate */
emilmont 1:fdd22bb7aa52 330 acc0 = __SMLALD(x2, y2, acc0);
emilmont 1:fdd22bb7aa52 331 acc2 = __SMLALD(x1, y2, acc2);
emilmont 1:fdd22bb7aa52 332
emilmont 1:fdd22bb7aa52 333 /* pack input data */
emilmont 1:fdd22bb7aa52 334 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 335 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 336 #else
emilmont 1:fdd22bb7aa52 337 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 338 #endif
emilmont 1:fdd22bb7aa52 339
emilmont 1:fdd22bb7aa52 340 acc3 = __SMLALDX(x3, y1, acc3);
emilmont 1:fdd22bb7aa52 341 acc1 = __SMLALDX(x3, y2, acc1);
emilmont 1:fdd22bb7aa52 342
emilmont 1:fdd22bb7aa52 343 x2 = _SIMD32_OFFSET(pScr1 + 2u);
emilmont 1:fdd22bb7aa52 344
emilmont 1:fdd22bb7aa52 345 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 346 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 347 #else
emilmont 1:fdd22bb7aa52 348 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 349 #endif
emilmont 1:fdd22bb7aa52 350
emilmont 1:fdd22bb7aa52 351 acc3 = __SMLALDX(x3, y2, acc3);
emilmont 1:fdd22bb7aa52 352
emilmont 1:fdd22bb7aa52 353 #else
emilmont 1:fdd22bb7aa52 354
emilmont 1:fdd22bb7aa52 355 /* Read four samples from smaller buffer */
emilmont 1:fdd22bb7aa52 356 a = *pIn2;
emilmont 1:fdd22bb7aa52 357 b = *(pIn2 + 1);
emilmont 1:fdd22bb7aa52 358
emilmont 1:fdd22bb7aa52 359 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 360 y1 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 361 #else
emilmont 1:fdd22bb7aa52 362 y1 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 363 #endif
emilmont 1:fdd22bb7aa52 364
emilmont 1:fdd22bb7aa52 365 a = *(pIn2 + 2);
emilmont 1:fdd22bb7aa52 366 b = *(pIn2 + 3);
emilmont 1:fdd22bb7aa52 367 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 368 y2 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 369 #else
emilmont 1:fdd22bb7aa52 370 y2 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 371 #endif
emilmont 1:fdd22bb7aa52 372
emilmont 1:fdd22bb7aa52 373 acc0 = __SMLALD(x1, y1, acc0);
emilmont 1:fdd22bb7aa52 374
emilmont 1:fdd22bb7aa52 375 acc2 = __SMLALD(x2, y1, acc2);
emilmont 1:fdd22bb7aa52 376
emilmont 1:fdd22bb7aa52 377 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 378 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 379 #else
emilmont 1:fdd22bb7aa52 380 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 381 #endif
emilmont 1:fdd22bb7aa52 382
emilmont 1:fdd22bb7aa52 383 acc1 = __SMLALDX(x3, y1, acc1);
emilmont 1:fdd22bb7aa52 384
emilmont 1:fdd22bb7aa52 385 a = *pScr1;
emilmont 1:fdd22bb7aa52 386 b = *(pScr1 + 1);
emilmont 1:fdd22bb7aa52 387
emilmont 1:fdd22bb7aa52 388 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 389 x1 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 390 #else
emilmont 1:fdd22bb7aa52 391 x1 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 392 #endif
emilmont 1:fdd22bb7aa52 393
emilmont 1:fdd22bb7aa52 394 acc0 = __SMLALD(x2, y2, acc0);
emilmont 1:fdd22bb7aa52 395
emilmont 1:fdd22bb7aa52 396 acc2 = __SMLALD(x1, y2, acc2);
emilmont 1:fdd22bb7aa52 397
emilmont 1:fdd22bb7aa52 398 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 399 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 400 #else
emilmont 1:fdd22bb7aa52 401 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 402 #endif
emilmont 1:fdd22bb7aa52 403
emilmont 1:fdd22bb7aa52 404 acc3 = __SMLALDX(x3, y1, acc3);
emilmont 1:fdd22bb7aa52 405
emilmont 1:fdd22bb7aa52 406 acc1 = __SMLALDX(x3, y2, acc1);
emilmont 1:fdd22bb7aa52 407
emilmont 1:fdd22bb7aa52 408 a = *(pScr1 + 2);
emilmont 1:fdd22bb7aa52 409 b = *(pScr1 + 3);
emilmont 1:fdd22bb7aa52 410
emilmont 1:fdd22bb7aa52 411 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 412 x2 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 413 #else
emilmont 1:fdd22bb7aa52 414 x2 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 415 #endif
emilmont 1:fdd22bb7aa52 416
emilmont 1:fdd22bb7aa52 417 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 418 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 419 #else
emilmont 1:fdd22bb7aa52 420 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 421 #endif
emilmont 1:fdd22bb7aa52 422
emilmont 1:fdd22bb7aa52 423 acc3 = __SMLALDX(x3, y2, acc3);
emilmont 1:fdd22bb7aa52 424
emilmont 1:fdd22bb7aa52 425 #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */
emilmont 1:fdd22bb7aa52 426
emilmont 1:fdd22bb7aa52 427 pIn2 += 4u;
emilmont 1:fdd22bb7aa52 428 pScr1 += 4u;
emilmont 1:fdd22bb7aa52 429
emilmont 1:fdd22bb7aa52 430
emilmont 1:fdd22bb7aa52 431 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 432 tapCnt--;
emilmont 1:fdd22bb7aa52 433 }
emilmont 1:fdd22bb7aa52 434
emilmont 1:fdd22bb7aa52 435 /* Update scratch pointer for remaining samples of smaller length sequence */
emilmont 1:fdd22bb7aa52 436 pScr1 -= 4u;
emilmont 1:fdd22bb7aa52 437
emilmont 1:fdd22bb7aa52 438 /* apply same above for remaining samples of smaller length sequence */
emilmont 1:fdd22bb7aa52 439 tapCnt = (srcBLen) & 3u;
emilmont 1:fdd22bb7aa52 440
emilmont 1:fdd22bb7aa52 441 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 442 {
emilmont 1:fdd22bb7aa52 443
emilmont 1:fdd22bb7aa52 444 /* accumlate the results */
emilmont 1:fdd22bb7aa52 445 acc0 += (*pScr1++ * *pIn2);
emilmont 1:fdd22bb7aa52 446 acc1 += (*pScr1++ * *pIn2);
emilmont 1:fdd22bb7aa52 447 acc2 += (*pScr1++ * *pIn2);
emilmont 1:fdd22bb7aa52 448 acc3 += (*pScr1++ * *pIn2++);
emilmont 1:fdd22bb7aa52 449
emilmont 1:fdd22bb7aa52 450 pScr1 -= 3u;
emilmont 1:fdd22bb7aa52 451
emilmont 1:fdd22bb7aa52 452 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 453 tapCnt--;
emilmont 1:fdd22bb7aa52 454 }
emilmont 1:fdd22bb7aa52 455
emilmont 1:fdd22bb7aa52 456 blkCnt--;
emilmont 1:fdd22bb7aa52 457
emilmont 1:fdd22bb7aa52 458
emilmont 1:fdd22bb7aa52 459 /* Store the results in the accumulators in the destination buffer. */
emilmont 1:fdd22bb7aa52 460
emilmont 1:fdd22bb7aa52 461 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 462
emilmont 1:fdd22bb7aa52 463 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 464 __PKHBT(__SSAT((acc0 >> 15), 16), __SSAT((acc1 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 465
emilmont 1:fdd22bb7aa52 466 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 467 __PKHBT(__SSAT((acc2 >> 15), 16), __SSAT((acc3 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 468
emilmont 1:fdd22bb7aa52 469 #else
emilmont 1:fdd22bb7aa52 470
emilmont 1:fdd22bb7aa52 471 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 472 __PKHBT(__SSAT((acc1 >> 15), 16), __SSAT((acc0 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 473
emilmont 1:fdd22bb7aa52 474 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 475 __PKHBT(__SSAT((acc3 >> 15), 16), __SSAT((acc2 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 476
emilmont 1:fdd22bb7aa52 477
emilmont 1:fdd22bb7aa52 478 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 479
emilmont 1:fdd22bb7aa52 480 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 481 pIn2 = py;
emilmont 1:fdd22bb7aa52 482
emilmont 1:fdd22bb7aa52 483 pScratch1 += 4u;
emilmont 1:fdd22bb7aa52 484
emilmont 1:fdd22bb7aa52 485 }
emilmont 1:fdd22bb7aa52 486
emilmont 1:fdd22bb7aa52 487
emilmont 1:fdd22bb7aa52 488 blkCnt = (srcALen + srcBLen - 1u) & 0x3;
emilmont 1:fdd22bb7aa52 489
emilmont 1:fdd22bb7aa52 490 /* Calculate convolution for remaining samples of Bigger length sequence */
emilmont 1:fdd22bb7aa52 491 while(blkCnt > 0)
emilmont 1:fdd22bb7aa52 492 {
emilmont 1:fdd22bb7aa52 493 /* Initialze temporary scratch pointer as scratch1 */
emilmont 1:fdd22bb7aa52 494 pScr1 = pScratch1;
emilmont 1:fdd22bb7aa52 495
emilmont 1:fdd22bb7aa52 496 /* Clear Accumlators */
emilmont 1:fdd22bb7aa52 497 acc0 = 0;
emilmont 1:fdd22bb7aa52 498
emilmont 1:fdd22bb7aa52 499 tapCnt = (srcBLen) >> 1u;
emilmont 1:fdd22bb7aa52 500
emilmont 1:fdd22bb7aa52 501 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 502 {
emilmont 1:fdd22bb7aa52 503
emilmont 1:fdd22bb7aa52 504 /* Read next two samples from scratch1 buffer */
emilmont 1:fdd22bb7aa52 505 acc0 += (*pScr1++ * *pIn2++);
emilmont 1:fdd22bb7aa52 506 acc0 += (*pScr1++ * *pIn2++);
emilmont 1:fdd22bb7aa52 507
emilmont 1:fdd22bb7aa52 508 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 509 tapCnt--;
emilmont 1:fdd22bb7aa52 510 }
emilmont 1:fdd22bb7aa52 511
emilmont 1:fdd22bb7aa52 512 tapCnt = (srcBLen) & 1u;
emilmont 1:fdd22bb7aa52 513
emilmont 1:fdd22bb7aa52 514 /* apply same above for remaining samples of smaller length sequence */
emilmont 1:fdd22bb7aa52 515 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 516 {
emilmont 1:fdd22bb7aa52 517
emilmont 1:fdd22bb7aa52 518 /* accumlate the results */
emilmont 1:fdd22bb7aa52 519 acc0 += (*pScr1++ * *pIn2++);
emilmont 1:fdd22bb7aa52 520
emilmont 1:fdd22bb7aa52 521 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 522 tapCnt--;
emilmont 1:fdd22bb7aa52 523 }
emilmont 1:fdd22bb7aa52 524
emilmont 1:fdd22bb7aa52 525 blkCnt--;
emilmont 1:fdd22bb7aa52 526
emilmont 1:fdd22bb7aa52 527 /* The result is in 2.30 format. Convert to 1.15 with saturation.
emilmont 1:fdd22bb7aa52 528 ** Then store the output in the destination buffer. */
emilmont 1:fdd22bb7aa52 529 *pOut++ = (q15_t) (__SSAT((acc0 >> 15), 16));
emilmont 1:fdd22bb7aa52 530
emilmont 1:fdd22bb7aa52 531
emilmont 1:fdd22bb7aa52 532 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 533 pIn2 = py;
emilmont 1:fdd22bb7aa52 534
emilmont 1:fdd22bb7aa52 535 pScratch1 += 1u;
emilmont 1:fdd22bb7aa52 536
emilmont 1:fdd22bb7aa52 537 }
emilmont 1:fdd22bb7aa52 538
emilmont 1:fdd22bb7aa52 539 }
emilmont 1:fdd22bb7aa52 540
emilmont 1:fdd22bb7aa52 541
emilmont 1:fdd22bb7aa52 542 /**
emilmont 1:fdd22bb7aa52 543 * @} end of Conv group
emilmont 1:fdd22bb7aa52 544 */