mbed official / mbed-dsp

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_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?

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_fast_opt_q15.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 1:fdd22bb7aa52 10 * Description: Fast Q15 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 Conv
emilmont 1:fdd22bb7aa52 44 * @{
emilmont 1:fdd22bb7aa52 45 */
emilmont 1:fdd22bb7aa52 46
emilmont 1:fdd22bb7aa52 47 /**
emilmont 1:fdd22bb7aa52 48 * @brief 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. Length srcALen+srcBLen-1.
emilmont 1:fdd22bb7aa52 54 * @param[in] *pScratch1 points to scratch buffer of size max(srcALen, srcBLen) + 2*min(srcALen, srcBLen) - 2.
emilmont 1:fdd22bb7aa52 55 * @param[in] *pScratch2 points to scratch buffer of size min(srcALen, srcBLen).
emilmont 1:fdd22bb7aa52 56 * @return none.
emilmont 1:fdd22bb7aa52 57 *
emilmont 1:fdd22bb7aa52 58 * \par Restrictions
emilmont 1:fdd22bb7aa52 59 * If the silicon does not support unaligned memory access enable the macro UNALIGNED_SUPPORT_DISABLE
emilmont 1:fdd22bb7aa52 60 * In this case input, output, scratch1 and scratch2 buffers should be aligned by 32-bit
emilmont 1:fdd22bb7aa52 61 *
emilmont 1:fdd22bb7aa52 62 * <b>Scaling and Overflow Behavior:</b>
emilmont 1:fdd22bb7aa52 63 *
emilmont 1:fdd22bb7aa52 64 * \par
emilmont 1:fdd22bb7aa52 65 * This fast version uses a 32-bit accumulator with 2.30 format.
emilmont 1:fdd22bb7aa52 66 * The accumulator maintains full precision of the intermediate multiplication results
emilmont 1:fdd22bb7aa52 67 * but provides only a single guard bit. There is no saturation on intermediate additions.
emilmont 1:fdd22bb7aa52 68 * Thus, if the accumulator overflows it wraps around and distorts the result.
emilmont 1:fdd22bb7aa52 69 * The input signals should be scaled down to avoid intermediate overflows.
emilmont 1:fdd22bb7aa52 70 * Scale down the inputs by log2(min(srcALen, srcBLen)) (log2 is read as log to the base 2) times to avoid overflows,
emilmont 1:fdd22bb7aa52 71 * as maximum of min(srcALen, srcBLen) number of additions are carried internally.
emilmont 1:fdd22bb7aa52 72 * The 2.30 accumulator is right shifted by 15 bits and then saturated to 1.15 format to yield the final result.
emilmont 1:fdd22bb7aa52 73 *
emilmont 1:fdd22bb7aa52 74 * \par
emilmont 1:fdd22bb7aa52 75 * See <code>arm_conv_q15()</code> for a slower implementation of this function which uses 64-bit accumulation to avoid wrap around distortion.
emilmont 1:fdd22bb7aa52 76 */
emilmont 1:fdd22bb7aa52 77
emilmont 1:fdd22bb7aa52 78 void arm_conv_fast_opt_q15(
emilmont 1:fdd22bb7aa52 79 q15_t * pSrcA,
emilmont 1:fdd22bb7aa52 80 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 81 q15_t * pSrcB,
emilmont 1:fdd22bb7aa52 82 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 83 q15_t * pDst,
emilmont 1:fdd22bb7aa52 84 q15_t * pScratch1,
emilmont 1:fdd22bb7aa52 85 q15_t * pScratch2)
emilmont 1:fdd22bb7aa52 86 {
emilmont 1:fdd22bb7aa52 87 q31_t acc0, acc1, acc2, acc3; /* Accumulators */
emilmont 1:fdd22bb7aa52 88 q31_t x1, x2, x3; /* Temporary variables to hold state and coefficient values */
emilmont 1:fdd22bb7aa52 89 q31_t y1, y2; /* State variables */
emilmont 1:fdd22bb7aa52 90 q15_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 91 q15_t *pScr1 = pScratch1; /* Temporary pointer for scratch1 */
emilmont 1:fdd22bb7aa52 92 q15_t *pScr2 = pScratch2; /* Temporary pointer for scratch1 */
emilmont 1:fdd22bb7aa52 93 q15_t *pIn1; /* inputA pointer */
emilmont 1:fdd22bb7aa52 94 q15_t *pIn2; /* inputB pointer */
emilmont 1:fdd22bb7aa52 95 q15_t *px; /* Intermediate inputA pointer */
emilmont 1:fdd22bb7aa52 96 q15_t *py; /* Intermediate inputB pointer */
emilmont 1:fdd22bb7aa52 97 uint32_t j, k, blkCnt; /* loop counter */
emilmont 1:fdd22bb7aa52 98 uint32_t tapCnt; /* loop count */
emilmont 1:fdd22bb7aa52 99 #ifdef UNALIGNED_SUPPORT_DISABLE
emilmont 1:fdd22bb7aa52 100
emilmont 1:fdd22bb7aa52 101 q15_t a, b;
emilmont 1:fdd22bb7aa52 102
emilmont 1:fdd22bb7aa52 103 #endif /* #ifdef UNALIGNED_SUPPORT_DISABLE */
emilmont 1:fdd22bb7aa52 104
emilmont 1:fdd22bb7aa52 105 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 106 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 107 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 108 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 109 {
emilmont 1:fdd22bb7aa52 110 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 111 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 112
emilmont 1:fdd22bb7aa52 113 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 114 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 115 }
emilmont 1:fdd22bb7aa52 116 else
emilmont 1:fdd22bb7aa52 117 {
emilmont 1:fdd22bb7aa52 118 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 119 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 120
emilmont 1:fdd22bb7aa52 121 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 122 pIn2 = pSrcA;
emilmont 1:fdd22bb7aa52 123
emilmont 1:fdd22bb7aa52 124 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 125 j = srcBLen;
emilmont 1:fdd22bb7aa52 126 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 127 srcALen = j;
emilmont 1:fdd22bb7aa52 128 }
emilmont 1:fdd22bb7aa52 129
emilmont 1:fdd22bb7aa52 130 /* Pointer to take end of scratch2 buffer */
emilmont 1:fdd22bb7aa52 131 pScr2 = pScratch2 + srcBLen - 1;
emilmont 1:fdd22bb7aa52 132
emilmont 1:fdd22bb7aa52 133 /* points to smaller length sequence */
emilmont 1:fdd22bb7aa52 134 px = pIn2;
emilmont 1:fdd22bb7aa52 135
emilmont 1:fdd22bb7aa52 136 /* Apply loop unrolling and do 4 Copies simultaneously. */
emilmont 1:fdd22bb7aa52 137 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 138
emilmont 1:fdd22bb7aa52 139 /* First part of the processing with loop unrolling copies 4 data points at a time.
emilmont 1:fdd22bb7aa52 140 ** a second loop below copies for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 141
emilmont 1:fdd22bb7aa52 142 /* Copy smaller length input sequence in reverse order into second scratch buffer */
emilmont 1:fdd22bb7aa52 143 while(k > 0u)
emilmont 1:fdd22bb7aa52 144 {
emilmont 1:fdd22bb7aa52 145 /* copy second buffer in reversal manner */
emilmont 1:fdd22bb7aa52 146 *pScr2-- = *px++;
emilmont 1:fdd22bb7aa52 147 *pScr2-- = *px++;
emilmont 1:fdd22bb7aa52 148 *pScr2-- = *px++;
emilmont 1:fdd22bb7aa52 149 *pScr2-- = *px++;
emilmont 1:fdd22bb7aa52 150
emilmont 1:fdd22bb7aa52 151 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 152 k--;
emilmont 1:fdd22bb7aa52 153 }
emilmont 1:fdd22bb7aa52 154
emilmont 1:fdd22bb7aa52 155 /* If the count is not a multiple of 4, copy remaining samples here.
emilmont 1:fdd22bb7aa52 156 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 157 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 158
emilmont 1:fdd22bb7aa52 159 while(k > 0u)
emilmont 1:fdd22bb7aa52 160 {
emilmont 1:fdd22bb7aa52 161 /* copy second buffer in reversal manner for remaining samples */
emilmont 1:fdd22bb7aa52 162 *pScr2-- = *px++;
emilmont 1:fdd22bb7aa52 163
emilmont 1:fdd22bb7aa52 164 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 165 k--;
emilmont 1:fdd22bb7aa52 166 }
emilmont 1:fdd22bb7aa52 167
emilmont 1:fdd22bb7aa52 168 /* Initialze temporary scratch pointer */
emilmont 1:fdd22bb7aa52 169 pScr1 = pScratch1;
emilmont 1:fdd22bb7aa52 170
emilmont 1:fdd22bb7aa52 171 /* Assuming scratch1 buffer is aligned by 32-bit */
emilmont 1:fdd22bb7aa52 172 /* Fill (srcBLen - 1u) zeros in scratch1 buffer */
emilmont 1:fdd22bb7aa52 173 arm_fill_q15(0, pScr1, (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 174
emilmont 1:fdd22bb7aa52 175 /* Update temporary scratch pointer */
emilmont 1:fdd22bb7aa52 176 pScr1 += (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 177
emilmont 1:fdd22bb7aa52 178 /* Copy bigger length sequence(srcALen) samples in scratch1 buffer */
emilmont 1:fdd22bb7aa52 179
emilmont 1:fdd22bb7aa52 180 #ifndef UNALIGNED_SUPPORT_DISABLE
emilmont 1:fdd22bb7aa52 181
emilmont 1:fdd22bb7aa52 182 /* Copy (srcALen) samples in scratch buffer */
emilmont 1:fdd22bb7aa52 183 arm_copy_q15(pIn1, pScr1, srcALen);
emilmont 1:fdd22bb7aa52 184
emilmont 1:fdd22bb7aa52 185 /* Update pointers */
emilmont 1:fdd22bb7aa52 186 pScr1 += srcALen;
emilmont 1:fdd22bb7aa52 187
emilmont 1:fdd22bb7aa52 188 #else
emilmont 1:fdd22bb7aa52 189
emilmont 1:fdd22bb7aa52 190 /* Apply loop unrolling and do 4 Copies simultaneously. */
emilmont 1:fdd22bb7aa52 191 k = srcALen >> 2u;
emilmont 1:fdd22bb7aa52 192
emilmont 1:fdd22bb7aa52 193 /* First part of the processing with loop unrolling copies 4 data points at a time.
emilmont 1:fdd22bb7aa52 194 ** a second loop below copies for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 195 while(k > 0u)
emilmont 1:fdd22bb7aa52 196 {
emilmont 1:fdd22bb7aa52 197 /* copy second buffer in reversal manner */
emilmont 1:fdd22bb7aa52 198 *pScr1++ = *pIn1++;
emilmont 1:fdd22bb7aa52 199 *pScr1++ = *pIn1++;
emilmont 1:fdd22bb7aa52 200 *pScr1++ = *pIn1++;
emilmont 1:fdd22bb7aa52 201 *pScr1++ = *pIn1++;
emilmont 1:fdd22bb7aa52 202
emilmont 1:fdd22bb7aa52 203 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 204 k--;
emilmont 1:fdd22bb7aa52 205 }
emilmont 1:fdd22bb7aa52 206
emilmont 1:fdd22bb7aa52 207 /* If the count is not a multiple of 4, copy remaining samples here.
emilmont 1:fdd22bb7aa52 208 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 209 k = srcALen % 0x4u;
emilmont 1:fdd22bb7aa52 210
emilmont 1:fdd22bb7aa52 211 while(k > 0u)
emilmont 1:fdd22bb7aa52 212 {
emilmont 1:fdd22bb7aa52 213 /* copy second buffer in reversal manner for remaining samples */
emilmont 1:fdd22bb7aa52 214 *pScr1++ = *pIn1++;
emilmont 1:fdd22bb7aa52 215
emilmont 1:fdd22bb7aa52 216 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 217 k--;
emilmont 1:fdd22bb7aa52 218 }
emilmont 1:fdd22bb7aa52 219
emilmont 1:fdd22bb7aa52 220 #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */
emilmont 1:fdd22bb7aa52 221
emilmont 1:fdd22bb7aa52 222
emilmont 1:fdd22bb7aa52 223 #ifndef UNALIGNED_SUPPORT_DISABLE
emilmont 1:fdd22bb7aa52 224
emilmont 1:fdd22bb7aa52 225 /* Fill (srcBLen - 1u) zeros at end of scratch buffer */
emilmont 1:fdd22bb7aa52 226 arm_fill_q15(0, pScr1, (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 227
emilmont 1:fdd22bb7aa52 228 /* Update pointer */
emilmont 1:fdd22bb7aa52 229 pScr1 += (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 230
emilmont 1:fdd22bb7aa52 231 #else
emilmont 1:fdd22bb7aa52 232
emilmont 1:fdd22bb7aa52 233 /* Apply loop unrolling and do 4 Copies simultaneously. */
emilmont 1:fdd22bb7aa52 234 k = (srcBLen - 1u) >> 2u;
emilmont 1:fdd22bb7aa52 235
emilmont 1:fdd22bb7aa52 236 /* First part of the processing with loop unrolling copies 4 data points at a time.
emilmont 1:fdd22bb7aa52 237 ** a second loop below copies for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 238 while(k > 0u)
emilmont 1:fdd22bb7aa52 239 {
emilmont 1:fdd22bb7aa52 240 /* copy second buffer in reversal manner */
emilmont 1:fdd22bb7aa52 241 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 242 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 243 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 244 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 245
emilmont 1:fdd22bb7aa52 246 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 247 k--;
emilmont 1:fdd22bb7aa52 248 }
emilmont 1:fdd22bb7aa52 249
emilmont 1:fdd22bb7aa52 250 /* If the count is not a multiple of 4, copy remaining samples here.
emilmont 1:fdd22bb7aa52 251 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 252 k = (srcBLen - 1u) % 0x4u;
emilmont 1:fdd22bb7aa52 253
emilmont 1:fdd22bb7aa52 254 while(k > 0u)
emilmont 1:fdd22bb7aa52 255 {
emilmont 1:fdd22bb7aa52 256 /* copy second buffer in reversal manner for remaining samples */
emilmont 1:fdd22bb7aa52 257 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 258
emilmont 1:fdd22bb7aa52 259 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 260 k--;
emilmont 1:fdd22bb7aa52 261 }
emilmont 1:fdd22bb7aa52 262
emilmont 1:fdd22bb7aa52 263 #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */
emilmont 1:fdd22bb7aa52 264
emilmont 1:fdd22bb7aa52 265 /* Temporary pointer for scratch2 */
emilmont 1:fdd22bb7aa52 266 py = pScratch2;
emilmont 1:fdd22bb7aa52 267
emilmont 1:fdd22bb7aa52 268
emilmont 1:fdd22bb7aa52 269 /* Initialization of pIn2 pointer */
emilmont 1:fdd22bb7aa52 270 pIn2 = py;
emilmont 1:fdd22bb7aa52 271
emilmont 1:fdd22bb7aa52 272 /* First part of the processing with loop unrolling process 4 data points at a time.
emilmont 1:fdd22bb7aa52 273 ** a second loop below process for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 274
emilmont 1:fdd22bb7aa52 275 /* Actual convolution process starts here */
emilmont 1:fdd22bb7aa52 276 blkCnt = (srcALen + srcBLen - 1u) >> 2;
emilmont 1:fdd22bb7aa52 277
emilmont 1:fdd22bb7aa52 278 while(blkCnt > 0)
emilmont 1:fdd22bb7aa52 279 {
emilmont 1:fdd22bb7aa52 280 /* Initialze temporary scratch pointer as scratch1 */
emilmont 1:fdd22bb7aa52 281 pScr1 = pScratch1;
emilmont 1:fdd22bb7aa52 282
emilmont 1:fdd22bb7aa52 283 /* Clear Accumlators */
emilmont 1:fdd22bb7aa52 284 acc0 = 0;
emilmont 1:fdd22bb7aa52 285 acc1 = 0;
emilmont 1:fdd22bb7aa52 286 acc2 = 0;
emilmont 1:fdd22bb7aa52 287 acc3 = 0;
emilmont 1:fdd22bb7aa52 288
emilmont 1:fdd22bb7aa52 289 /* Read two samples from scratch1 buffer */
emilmont 1:fdd22bb7aa52 290 x1 = *__SIMD32(pScr1)++;
emilmont 1:fdd22bb7aa52 291
emilmont 1:fdd22bb7aa52 292 /* Read next two samples from scratch1 buffer */
emilmont 1:fdd22bb7aa52 293 x2 = *__SIMD32(pScr1)++;
emilmont 1:fdd22bb7aa52 294
emilmont 1:fdd22bb7aa52 295 tapCnt = (srcBLen) >> 2u;
emilmont 1:fdd22bb7aa52 296
emilmont 1:fdd22bb7aa52 297 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 298 {
emilmont 1:fdd22bb7aa52 299
emilmont 1:fdd22bb7aa52 300 #ifndef UNALIGNED_SUPPORT_DISABLE
emilmont 1:fdd22bb7aa52 301
emilmont 1:fdd22bb7aa52 302 /* Read four samples from smaller buffer */
emilmont 1:fdd22bb7aa52 303 y1 = _SIMD32_OFFSET(pIn2);
emilmont 1:fdd22bb7aa52 304 y2 = _SIMD32_OFFSET(pIn2 + 2u);
emilmont 1:fdd22bb7aa52 305
emilmont 1:fdd22bb7aa52 306 /* multiply and accumlate */
emilmont 1:fdd22bb7aa52 307 acc0 = __SMLAD(x1, y1, acc0);
emilmont 1:fdd22bb7aa52 308 acc2 = __SMLAD(x2, y1, acc2);
emilmont 1:fdd22bb7aa52 309
emilmont 1:fdd22bb7aa52 310 /* pack input data */
emilmont 1:fdd22bb7aa52 311 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 312 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 313 #else
emilmont 1:fdd22bb7aa52 314 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 315 #endif
emilmont 1:fdd22bb7aa52 316
emilmont 1:fdd22bb7aa52 317 /* multiply and accumlate */
emilmont 1:fdd22bb7aa52 318 acc1 = __SMLADX(x3, y1, acc1);
emilmont 1:fdd22bb7aa52 319
emilmont 1:fdd22bb7aa52 320 /* Read next two samples from scratch1 buffer */
emilmont 1:fdd22bb7aa52 321 x1 = _SIMD32_OFFSET(pScr1);
emilmont 1:fdd22bb7aa52 322
emilmont 1:fdd22bb7aa52 323 /* multiply and accumlate */
emilmont 1:fdd22bb7aa52 324 acc0 = __SMLAD(x2, y2, acc0);
emilmont 1:fdd22bb7aa52 325 acc2 = __SMLAD(x1, y2, acc2);
emilmont 1:fdd22bb7aa52 326
emilmont 1:fdd22bb7aa52 327 /* pack input data */
emilmont 1:fdd22bb7aa52 328 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 329 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 330 #else
emilmont 1:fdd22bb7aa52 331 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 332 #endif
emilmont 1:fdd22bb7aa52 333
emilmont 1:fdd22bb7aa52 334 acc3 = __SMLADX(x3, y1, acc3);
emilmont 1:fdd22bb7aa52 335 acc1 = __SMLADX(x3, y2, acc1);
emilmont 1:fdd22bb7aa52 336
emilmont 1:fdd22bb7aa52 337 x2 = _SIMD32_OFFSET(pScr1 + 2u);
emilmont 1:fdd22bb7aa52 338
emilmont 1:fdd22bb7aa52 339 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 340 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 341 #else
emilmont 1:fdd22bb7aa52 342 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 343 #endif
emilmont 1:fdd22bb7aa52 344
emilmont 1:fdd22bb7aa52 345 acc3 = __SMLADX(x3, y2, acc3);
emilmont 1:fdd22bb7aa52 346
emilmont 1:fdd22bb7aa52 347 #else
emilmont 1:fdd22bb7aa52 348
emilmont 1:fdd22bb7aa52 349 /* Read four samples from smaller buffer */
emilmont 1:fdd22bb7aa52 350 a = *pIn2;
emilmont 1:fdd22bb7aa52 351 b = *(pIn2 + 1);
emilmont 1:fdd22bb7aa52 352
emilmont 1:fdd22bb7aa52 353 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 354 y1 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 355 #else
emilmont 1:fdd22bb7aa52 356 y1 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 357 #endif
emilmont 1:fdd22bb7aa52 358
emilmont 1:fdd22bb7aa52 359 a = *(pIn2 + 2);
emilmont 1:fdd22bb7aa52 360 b = *(pIn2 + 3);
emilmont 1:fdd22bb7aa52 361 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 362 y2 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 363 #else
emilmont 1:fdd22bb7aa52 364 y2 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 365 #endif
emilmont 1:fdd22bb7aa52 366
emilmont 1:fdd22bb7aa52 367 acc0 = __SMLAD(x1, y1, acc0);
emilmont 1:fdd22bb7aa52 368
emilmont 1:fdd22bb7aa52 369 acc2 = __SMLAD(x2, y1, acc2);
emilmont 1:fdd22bb7aa52 370
emilmont 1:fdd22bb7aa52 371 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 372 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 373 #else
emilmont 1:fdd22bb7aa52 374 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 375 #endif
emilmont 1:fdd22bb7aa52 376
emilmont 1:fdd22bb7aa52 377 acc1 = __SMLADX(x3, y1, acc1);
emilmont 1:fdd22bb7aa52 378
emilmont 1:fdd22bb7aa52 379 a = *pScr1;
emilmont 1:fdd22bb7aa52 380 b = *(pScr1 + 1);
emilmont 1:fdd22bb7aa52 381
emilmont 1:fdd22bb7aa52 382 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 383 x1 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 384 #else
emilmont 1:fdd22bb7aa52 385 x1 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 386 #endif
emilmont 1:fdd22bb7aa52 387
emilmont 1:fdd22bb7aa52 388 acc0 = __SMLAD(x2, y2, acc0);
emilmont 1:fdd22bb7aa52 389
emilmont 1:fdd22bb7aa52 390 acc2 = __SMLAD(x1, y2, acc2);
emilmont 1:fdd22bb7aa52 391
emilmont 1:fdd22bb7aa52 392 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 393 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 394 #else
emilmont 1:fdd22bb7aa52 395 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 396 #endif
emilmont 1:fdd22bb7aa52 397
emilmont 1:fdd22bb7aa52 398 acc3 = __SMLADX(x3, y1, acc3);
emilmont 1:fdd22bb7aa52 399
emilmont 1:fdd22bb7aa52 400 acc1 = __SMLADX(x3, y2, acc1);
emilmont 1:fdd22bb7aa52 401
emilmont 1:fdd22bb7aa52 402 a = *(pScr1 + 2);
emilmont 1:fdd22bb7aa52 403 b = *(pScr1 + 3);
emilmont 1:fdd22bb7aa52 404
emilmont 1:fdd22bb7aa52 405 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 406 x2 = __PKHBT(a, b, 16);
emilmont 1:fdd22bb7aa52 407 #else
emilmont 1:fdd22bb7aa52 408 x2 = __PKHBT(b, a, 16);
emilmont 1:fdd22bb7aa52 409 #endif
emilmont 1:fdd22bb7aa52 410
emilmont 1:fdd22bb7aa52 411 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 412 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 413 #else
emilmont 1:fdd22bb7aa52 414 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 415 #endif
emilmont 1:fdd22bb7aa52 416
emilmont 1:fdd22bb7aa52 417 acc3 = __SMLADX(x3, y2, acc3);
emilmont 1:fdd22bb7aa52 418
emilmont 1:fdd22bb7aa52 419 #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */
emilmont 1:fdd22bb7aa52 420
emilmont 1:fdd22bb7aa52 421 /* update scratch pointers */
emilmont 1:fdd22bb7aa52 422 pIn2 += 4u;
emilmont 1:fdd22bb7aa52 423 pScr1 += 4u;
emilmont 1:fdd22bb7aa52 424
emilmont 1:fdd22bb7aa52 425
emilmont 1:fdd22bb7aa52 426 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 427 tapCnt--;
emilmont 1:fdd22bb7aa52 428 }
emilmont 1:fdd22bb7aa52 429
emilmont 1:fdd22bb7aa52 430 /* Update scratch pointer for remaining samples of smaller length sequence */
emilmont 1:fdd22bb7aa52 431 pScr1 -= 4u;
emilmont 1:fdd22bb7aa52 432
emilmont 1:fdd22bb7aa52 433 /* apply same above for remaining samples of smaller length sequence */
emilmont 1:fdd22bb7aa52 434 tapCnt = (srcBLen) & 3u;
emilmont 1:fdd22bb7aa52 435
emilmont 1:fdd22bb7aa52 436 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 437 {
emilmont 1:fdd22bb7aa52 438
emilmont 1:fdd22bb7aa52 439 /* accumlate the results */
emilmont 1:fdd22bb7aa52 440 acc0 += (*pScr1++ * *pIn2);
emilmont 1:fdd22bb7aa52 441 acc1 += (*pScr1++ * *pIn2);
emilmont 1:fdd22bb7aa52 442 acc2 += (*pScr1++ * *pIn2);
emilmont 1:fdd22bb7aa52 443 acc3 += (*pScr1++ * *pIn2++);
emilmont 1:fdd22bb7aa52 444
emilmont 1:fdd22bb7aa52 445 pScr1 -= 3u;
emilmont 1:fdd22bb7aa52 446
emilmont 1:fdd22bb7aa52 447 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 448 tapCnt--;
emilmont 1:fdd22bb7aa52 449 }
emilmont 1:fdd22bb7aa52 450
emilmont 1:fdd22bb7aa52 451 blkCnt--;
emilmont 1:fdd22bb7aa52 452
emilmont 1:fdd22bb7aa52 453
emilmont 1:fdd22bb7aa52 454 /* Store the results in the accumulators in the destination buffer. */
emilmont 1:fdd22bb7aa52 455
emilmont 1:fdd22bb7aa52 456 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 457
emilmont 1:fdd22bb7aa52 458 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 459 __PKHBT(__SSAT((acc0 >> 15), 16), __SSAT((acc1 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 460
emilmont 1:fdd22bb7aa52 461 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 462 __PKHBT(__SSAT((acc2 >> 15), 16), __SSAT((acc3 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 463
emilmont 1:fdd22bb7aa52 464
emilmont 1:fdd22bb7aa52 465 #else
emilmont 1:fdd22bb7aa52 466
emilmont 1:fdd22bb7aa52 467 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 468 __PKHBT(__SSAT((acc1 >> 15), 16), __SSAT((acc0 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 469
emilmont 1:fdd22bb7aa52 470 *__SIMD32(pOut)++ =
emilmont 1:fdd22bb7aa52 471 __PKHBT(__SSAT((acc3 >> 15), 16), __SSAT((acc2 >> 15), 16), 16);
emilmont 1:fdd22bb7aa52 472
emilmont 1:fdd22bb7aa52 473
emilmont 1:fdd22bb7aa52 474
emilmont 1:fdd22bb7aa52 475 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
emilmont 1:fdd22bb7aa52 476
emilmont 1:fdd22bb7aa52 477 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 478 pIn2 = py;
emilmont 1:fdd22bb7aa52 479
emilmont 1:fdd22bb7aa52 480 pScratch1 += 4u;
emilmont 1:fdd22bb7aa52 481
emilmont 1:fdd22bb7aa52 482 }
emilmont 1:fdd22bb7aa52 483
emilmont 1:fdd22bb7aa52 484
emilmont 1:fdd22bb7aa52 485 blkCnt = (srcALen + srcBLen - 1u) & 0x3;
emilmont 1:fdd22bb7aa52 486
emilmont 1:fdd22bb7aa52 487 /* Calculate convolution for remaining samples of Bigger length sequence */
emilmont 1:fdd22bb7aa52 488 while(blkCnt > 0)
emilmont 1:fdd22bb7aa52 489 {
emilmont 1:fdd22bb7aa52 490 /* Initialze temporary scratch pointer as scratch1 */
emilmont 1:fdd22bb7aa52 491 pScr1 = pScratch1;
emilmont 1:fdd22bb7aa52 492
emilmont 1:fdd22bb7aa52 493 /* Clear Accumlators */
emilmont 1:fdd22bb7aa52 494 acc0 = 0;
emilmont 1:fdd22bb7aa52 495
emilmont 1:fdd22bb7aa52 496 tapCnt = (srcBLen) >> 1u;
emilmont 1:fdd22bb7aa52 497
emilmont 1:fdd22bb7aa52 498 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 499 {
emilmont 1:fdd22bb7aa52 500
emilmont 1:fdd22bb7aa52 501 acc0 += (*pScr1++ * *pIn2++);
emilmont 1:fdd22bb7aa52 502 acc0 += (*pScr1++ * *pIn2++);
emilmont 1:fdd22bb7aa52 503
emilmont 1:fdd22bb7aa52 504 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 505 tapCnt--;
emilmont 1:fdd22bb7aa52 506 }
emilmont 1:fdd22bb7aa52 507
emilmont 1:fdd22bb7aa52 508 tapCnt = (srcBLen) & 1u;
emilmont 1:fdd22bb7aa52 509
emilmont 1:fdd22bb7aa52 510 /* apply same above for remaining samples of smaller length sequence */
emilmont 1:fdd22bb7aa52 511 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 512 {
emilmont 1:fdd22bb7aa52 513
emilmont 1:fdd22bb7aa52 514 /* accumlate the results */
emilmont 1:fdd22bb7aa52 515 acc0 += (*pScr1++ * *pIn2++);
emilmont 1:fdd22bb7aa52 516
emilmont 1:fdd22bb7aa52 517 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 518 tapCnt--;
emilmont 1:fdd22bb7aa52 519 }
emilmont 1:fdd22bb7aa52 520
emilmont 1:fdd22bb7aa52 521 blkCnt--;
emilmont 1:fdd22bb7aa52 522
emilmont 1:fdd22bb7aa52 523 /* The result is in 2.30 format. Convert to 1.15 with saturation.
emilmont 1:fdd22bb7aa52 524 ** Then store the output in the destination buffer. */
emilmont 1:fdd22bb7aa52 525 *pOut++ = (q15_t) (__SSAT((acc0 >> 15), 16));
emilmont 1:fdd22bb7aa52 526
emilmont 1:fdd22bb7aa52 527 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 528 pIn2 = py;
emilmont 1:fdd22bb7aa52 529
emilmont 1:fdd22bb7aa52 530 pScratch1 += 1u;
emilmont 1:fdd22bb7aa52 531
emilmont 1:fdd22bb7aa52 532 }
emilmont 1:fdd22bb7aa52 533
emilmont 1:fdd22bb7aa52 534 }
emilmont 1:fdd22bb7aa52 535
emilmont 1:fdd22bb7aa52 536 /**
emilmont 1:fdd22bb7aa52 537 * @} end of Conv group
emilmont 1:fdd22bb7aa52 538 */