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_opt_q7.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_q7.c
emilmont 1:fdd22bb7aa52 9 *
emilmont 1:fdd22bb7aa52 10 * Description: Convolution of Q7 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 Q7 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 type q15_t) of size max(srcALen, srcBLen) + 2*min(srcALen, srcBLen) - 2.
emilmont 1:fdd22bb7aa52 59 * @param[in] *pScratch2 points to scratch buffer (of type q15_t) 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 * @details
emilmont 1:fdd22bb7aa52 67 * <b>Scaling and Overflow Behavior:</b>
emilmont 1:fdd22bb7aa52 68 *
emilmont 1:fdd22bb7aa52 69 * \par
emilmont 1:fdd22bb7aa52 70 * The function is implemented using a 32-bit internal accumulator.
emilmont 1:fdd22bb7aa52 71 * Both the inputs are represented in 1.7 format and multiplications yield a 2.14 result.
emilmont 1:fdd22bb7aa52 72 * The 2.14 intermediate results are accumulated in a 32-bit accumulator in 18.14 format.
emilmont 1:fdd22bb7aa52 73 * This approach provides 17 guard bits and there is no risk of overflow as long as <code>max(srcALen, srcBLen)<131072</code>.
emilmont 1:fdd22bb7aa52 74 * The 18.14 result is then truncated to 18.7 format by discarding the low 7 bits and then saturated to 1.7 format.
emilmont 1:fdd22bb7aa52 75 *
emilmont 1:fdd22bb7aa52 76 */
emilmont 1:fdd22bb7aa52 77
emilmont 1:fdd22bb7aa52 78 void arm_conv_opt_q7(
emilmont 1:fdd22bb7aa52 79 q7_t * pSrcA,
emilmont 1:fdd22bb7aa52 80 uint32_t srcALen,
emilmont 1:fdd22bb7aa52 81 q7_t * pSrcB,
emilmont 1:fdd22bb7aa52 82 uint32_t srcBLen,
emilmont 1:fdd22bb7aa52 83 q7_t * pDst,
emilmont 1:fdd22bb7aa52 84 q15_t * pScratch1,
emilmont 1:fdd22bb7aa52 85 q15_t * pScratch2)
emilmont 1:fdd22bb7aa52 86 {
emilmont 1:fdd22bb7aa52 87
emilmont 1:fdd22bb7aa52 88 q15_t *pScr2, *pScr1; /* Intermediate pointers for scratch pointers */
emilmont 1:fdd22bb7aa52 89 q15_t x4; /* Temporary input variable */
emilmont 1:fdd22bb7aa52 90 q7_t *pIn1, *pIn2; /* inputA and inputB pointer */
emilmont 1:fdd22bb7aa52 91 uint32_t j, k, blkCnt, tapCnt; /* loop counter */
emilmont 1:fdd22bb7aa52 92 q7_t *px; /* Temporary input1 pointer */
emilmont 1:fdd22bb7aa52 93 q15_t *py; /* Temporary input2 pointer */
emilmont 1:fdd22bb7aa52 94 q31_t acc0, acc1, acc2, acc3; /* Accumulator */
emilmont 1:fdd22bb7aa52 95 q31_t x1, x2, x3, y1; /* Temporary input variables */
emilmont 1:fdd22bb7aa52 96 q7_t *pOut = pDst; /* output pointer */
emilmont 1:fdd22bb7aa52 97 q7_t out0, out1, out2, out3; /* temporary variables */
emilmont 1:fdd22bb7aa52 98
emilmont 1:fdd22bb7aa52 99 /* The algorithm implementation is based on the lengths of the inputs. */
emilmont 1:fdd22bb7aa52 100 /* srcB is always made to slide across srcA. */
emilmont 1:fdd22bb7aa52 101 /* So srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 102 if(srcALen >= srcBLen)
emilmont 1:fdd22bb7aa52 103 {
emilmont 1:fdd22bb7aa52 104 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 105 pIn1 = pSrcA;
emilmont 1:fdd22bb7aa52 106
emilmont 1:fdd22bb7aa52 107 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 108 pIn2 = pSrcB;
emilmont 1:fdd22bb7aa52 109 }
emilmont 1:fdd22bb7aa52 110 else
emilmont 1:fdd22bb7aa52 111 {
emilmont 1:fdd22bb7aa52 112 /* Initialization of inputA pointer */
emilmont 1:fdd22bb7aa52 113 pIn1 = pSrcB;
emilmont 1:fdd22bb7aa52 114
emilmont 1:fdd22bb7aa52 115 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 116 pIn2 = pSrcA;
emilmont 1:fdd22bb7aa52 117
emilmont 1:fdd22bb7aa52 118 /* srcBLen is always considered as shorter or equal to srcALen */
emilmont 1:fdd22bb7aa52 119 j = srcBLen;
emilmont 1:fdd22bb7aa52 120 srcBLen = srcALen;
emilmont 1:fdd22bb7aa52 121 srcALen = j;
emilmont 1:fdd22bb7aa52 122 }
emilmont 1:fdd22bb7aa52 123
emilmont 1:fdd22bb7aa52 124 /* pointer to take end of scratch2 buffer */
emilmont 1:fdd22bb7aa52 125 pScr2 = pScratch2;
emilmont 1:fdd22bb7aa52 126
emilmont 1:fdd22bb7aa52 127 /* points to smaller length sequence */
emilmont 1:fdd22bb7aa52 128 px = pIn2 + srcBLen - 1;
emilmont 1:fdd22bb7aa52 129
emilmont 1:fdd22bb7aa52 130 /* Apply loop unrolling and do 4 Copies simultaneously. */
emilmont 1:fdd22bb7aa52 131 k = srcBLen >> 2u;
emilmont 1:fdd22bb7aa52 132
emilmont 1:fdd22bb7aa52 133 /* First part of the processing with loop unrolling copies 4 data points at a time.
emilmont 1:fdd22bb7aa52 134 ** a second loop below copies for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 135 while(k > 0u)
emilmont 1:fdd22bb7aa52 136 {
emilmont 1:fdd22bb7aa52 137 /* copy second buffer in reversal manner */
emilmont 1:fdd22bb7aa52 138 x4 = (q15_t) * px--;
emilmont 1:fdd22bb7aa52 139 *pScr2++ = x4;
emilmont 1:fdd22bb7aa52 140 x4 = (q15_t) * px--;
emilmont 1:fdd22bb7aa52 141 *pScr2++ = x4;
emilmont 1:fdd22bb7aa52 142 x4 = (q15_t) * px--;
emilmont 1:fdd22bb7aa52 143 *pScr2++ = x4;
emilmont 1:fdd22bb7aa52 144 x4 = (q15_t) * px--;
emilmont 1:fdd22bb7aa52 145 *pScr2++ = x4;
emilmont 1:fdd22bb7aa52 146
emilmont 1:fdd22bb7aa52 147 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 148 k--;
emilmont 1:fdd22bb7aa52 149 }
emilmont 1:fdd22bb7aa52 150
emilmont 1:fdd22bb7aa52 151 /* If the count is not a multiple of 4, copy remaining samples here.
emilmont 1:fdd22bb7aa52 152 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 153 k = srcBLen % 0x4u;
emilmont 1:fdd22bb7aa52 154
emilmont 1:fdd22bb7aa52 155 while(k > 0u)
emilmont 1:fdd22bb7aa52 156 {
emilmont 1:fdd22bb7aa52 157 /* copy second buffer in reversal manner for remaining samples */
emilmont 1:fdd22bb7aa52 158 x4 = (q15_t) * px--;
emilmont 1:fdd22bb7aa52 159 *pScr2++ = x4;
emilmont 1:fdd22bb7aa52 160
emilmont 1:fdd22bb7aa52 161 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 162 k--;
emilmont 1:fdd22bb7aa52 163 }
emilmont 1:fdd22bb7aa52 164
emilmont 1:fdd22bb7aa52 165 /* Initialze temporary scratch pointer */
emilmont 1:fdd22bb7aa52 166 pScr1 = pScratch1;
emilmont 1:fdd22bb7aa52 167
emilmont 1:fdd22bb7aa52 168 /* Fill (srcBLen - 1u) zeros in scratch buffer */
emilmont 1:fdd22bb7aa52 169 arm_fill_q15(0, pScr1, (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 170
emilmont 1:fdd22bb7aa52 171 /* Update temporary scratch pointer */
emilmont 1:fdd22bb7aa52 172 pScr1 += (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 173
emilmont 1:fdd22bb7aa52 174 /* Copy (srcALen) samples in scratch buffer */
emilmont 1:fdd22bb7aa52 175 /* Apply loop unrolling and do 4 Copies simultaneously. */
emilmont 1:fdd22bb7aa52 176 k = srcALen >> 2u;
emilmont 1:fdd22bb7aa52 177
emilmont 1:fdd22bb7aa52 178 /* First part of the processing with loop unrolling copies 4 data points at a time.
emilmont 1:fdd22bb7aa52 179 ** a second loop below copies for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 180 while(k > 0u)
emilmont 1:fdd22bb7aa52 181 {
emilmont 1:fdd22bb7aa52 182 /* copy second buffer in reversal manner */
emilmont 1:fdd22bb7aa52 183 x4 = (q15_t) * pIn1++;
emilmont 1:fdd22bb7aa52 184 *pScr1++ = x4;
emilmont 1:fdd22bb7aa52 185 x4 = (q15_t) * pIn1++;
emilmont 1:fdd22bb7aa52 186 *pScr1++ = x4;
emilmont 1:fdd22bb7aa52 187 x4 = (q15_t) * pIn1++;
emilmont 1:fdd22bb7aa52 188 *pScr1++ = x4;
emilmont 1:fdd22bb7aa52 189 x4 = (q15_t) * pIn1++;
emilmont 1:fdd22bb7aa52 190 *pScr1++ = x4;
emilmont 1:fdd22bb7aa52 191
emilmont 1:fdd22bb7aa52 192 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 193 k--;
emilmont 1:fdd22bb7aa52 194 }
emilmont 1:fdd22bb7aa52 195
emilmont 1:fdd22bb7aa52 196 /* If the count is not a multiple of 4, copy remaining samples here.
emilmont 1:fdd22bb7aa52 197 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 198 k = srcALen % 0x4u;
emilmont 1:fdd22bb7aa52 199
emilmont 1:fdd22bb7aa52 200 while(k > 0u)
emilmont 1:fdd22bb7aa52 201 {
emilmont 1:fdd22bb7aa52 202 /* copy second buffer in reversal manner for remaining samples */
emilmont 1:fdd22bb7aa52 203 x4 = (q15_t) * pIn1++;
emilmont 1:fdd22bb7aa52 204 *pScr1++ = x4;
emilmont 1:fdd22bb7aa52 205
emilmont 1:fdd22bb7aa52 206 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 207 k--;
emilmont 1:fdd22bb7aa52 208 }
emilmont 1:fdd22bb7aa52 209
emilmont 1:fdd22bb7aa52 210 #ifndef UNALIGNED_SUPPORT_DISABLE
emilmont 1:fdd22bb7aa52 211
emilmont 1:fdd22bb7aa52 212 /* Fill (srcBLen - 1u) zeros at end of scratch buffer */
emilmont 1:fdd22bb7aa52 213 arm_fill_q15(0, pScr1, (srcBLen - 1u));
emilmont 1:fdd22bb7aa52 214
emilmont 1:fdd22bb7aa52 215 /* Update pointer */
emilmont 1:fdd22bb7aa52 216 pScr1 += (srcBLen - 1u);
emilmont 1:fdd22bb7aa52 217
emilmont 1:fdd22bb7aa52 218 #else
emilmont 1:fdd22bb7aa52 219
emilmont 1:fdd22bb7aa52 220 /* Apply loop unrolling and do 4 Copies simultaneously. */
emilmont 1:fdd22bb7aa52 221 k = (srcBLen - 1u) >> 2u;
emilmont 1:fdd22bb7aa52 222
emilmont 1:fdd22bb7aa52 223 /* First part of the processing with loop unrolling copies 4 data points at a time.
emilmont 1:fdd22bb7aa52 224 ** a second loop below copies for the remaining 1 to 3 samples. */
emilmont 1:fdd22bb7aa52 225 while(k > 0u)
emilmont 1:fdd22bb7aa52 226 {
emilmont 1:fdd22bb7aa52 227 /* copy second buffer in reversal manner */
emilmont 1:fdd22bb7aa52 228 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 229 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 230 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 231 *pScr1++ = 0;
emilmont 1:fdd22bb7aa52 232
emilmont 1:fdd22bb7aa52 233 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 234 k--;
emilmont 1:fdd22bb7aa52 235 }
emilmont 1:fdd22bb7aa52 236
emilmont 1:fdd22bb7aa52 237 /* If the count is not a multiple of 4, copy remaining samples here.
emilmont 1:fdd22bb7aa52 238 ** No loop unrolling is used. */
emilmont 1:fdd22bb7aa52 239 k = (srcBLen - 1u) % 0x4u;
emilmont 1:fdd22bb7aa52 240
emilmont 1:fdd22bb7aa52 241 while(k > 0u)
emilmont 1:fdd22bb7aa52 242 {
emilmont 1:fdd22bb7aa52 243 /* copy second buffer in reversal manner for remaining samples */
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 #endif
emilmont 1:fdd22bb7aa52 251
emilmont 1:fdd22bb7aa52 252 /* Temporary pointer for scratch2 */
emilmont 1:fdd22bb7aa52 253 py = pScratch2;
emilmont 1:fdd22bb7aa52 254
emilmont 1:fdd22bb7aa52 255 /* Initialization of pIn2 pointer */
emilmont 1:fdd22bb7aa52 256 pIn2 = (q7_t *) py;
emilmont 1:fdd22bb7aa52 257
emilmont 1:fdd22bb7aa52 258 pScr2 = py;
emilmont 1:fdd22bb7aa52 259
emilmont 1:fdd22bb7aa52 260 /* Actual convolution process starts here */
emilmont 1:fdd22bb7aa52 261 blkCnt = (srcALen + srcBLen - 1u) >> 2;
emilmont 1:fdd22bb7aa52 262
emilmont 1:fdd22bb7aa52 263 while(blkCnt > 0)
emilmont 1:fdd22bb7aa52 264 {
emilmont 1:fdd22bb7aa52 265 /* Initialze temporary scratch pointer as scratch1 */
emilmont 1:fdd22bb7aa52 266 pScr1 = pScratch1;
emilmont 1:fdd22bb7aa52 267
emilmont 1:fdd22bb7aa52 268 /* Clear Accumlators */
emilmont 1:fdd22bb7aa52 269 acc0 = 0;
emilmont 1:fdd22bb7aa52 270 acc1 = 0;
emilmont 1:fdd22bb7aa52 271 acc2 = 0;
emilmont 1:fdd22bb7aa52 272 acc3 = 0;
emilmont 1:fdd22bb7aa52 273
emilmont 1:fdd22bb7aa52 274 /* Read two samples from scratch1 buffer */
emilmont 1:fdd22bb7aa52 275 x1 = *__SIMD32(pScr1)++;
emilmont 1:fdd22bb7aa52 276
emilmont 1:fdd22bb7aa52 277 /* Read next two samples from scratch1 buffer */
emilmont 1:fdd22bb7aa52 278 x2 = *__SIMD32(pScr1)++;
emilmont 1:fdd22bb7aa52 279
emilmont 1:fdd22bb7aa52 280 tapCnt = (srcBLen) >> 2u;
emilmont 1:fdd22bb7aa52 281
emilmont 1:fdd22bb7aa52 282 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 283 {
emilmont 1:fdd22bb7aa52 284
emilmont 1:fdd22bb7aa52 285 /* Read four samples from smaller buffer */
emilmont 1:fdd22bb7aa52 286 y1 = _SIMD32_OFFSET(pScr2);
emilmont 1:fdd22bb7aa52 287
emilmont 1:fdd22bb7aa52 288 /* multiply and accumlate */
emilmont 1:fdd22bb7aa52 289 acc0 = __SMLAD(x1, y1, acc0);
emilmont 1:fdd22bb7aa52 290 acc2 = __SMLAD(x2, y1, acc2);
emilmont 1:fdd22bb7aa52 291
emilmont 1:fdd22bb7aa52 292 /* pack input data */
emilmont 1:fdd22bb7aa52 293 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 294 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 295 #else
emilmont 1:fdd22bb7aa52 296 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 297 #endif
emilmont 1:fdd22bb7aa52 298
emilmont 1:fdd22bb7aa52 299 /* multiply and accumlate */
emilmont 1:fdd22bb7aa52 300 acc1 = __SMLADX(x3, y1, acc1);
emilmont 1:fdd22bb7aa52 301
emilmont 1:fdd22bb7aa52 302 /* Read next two samples from scratch1 buffer */
emilmont 1:fdd22bb7aa52 303 x1 = *__SIMD32(pScr1)++;
emilmont 1:fdd22bb7aa52 304
emilmont 1:fdd22bb7aa52 305 /* pack input data */
emilmont 1:fdd22bb7aa52 306 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 307 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 308 #else
emilmont 1:fdd22bb7aa52 309 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 310 #endif
emilmont 1:fdd22bb7aa52 311
emilmont 1:fdd22bb7aa52 312 acc3 = __SMLADX(x3, y1, acc3);
emilmont 1:fdd22bb7aa52 313
emilmont 1:fdd22bb7aa52 314 /* Read four samples from smaller buffer */
emilmont 1:fdd22bb7aa52 315 y1 = _SIMD32_OFFSET(pScr2 + 2u);
emilmont 1:fdd22bb7aa52 316
emilmont 1:fdd22bb7aa52 317 acc0 = __SMLAD(x2, y1, acc0);
emilmont 1:fdd22bb7aa52 318
emilmont 1:fdd22bb7aa52 319 acc2 = __SMLAD(x1, y1, acc2);
emilmont 1:fdd22bb7aa52 320
emilmont 1:fdd22bb7aa52 321 acc1 = __SMLADX(x3, y1, acc1);
emilmont 1:fdd22bb7aa52 322
emilmont 1:fdd22bb7aa52 323 x2 = *__SIMD32(pScr1)++;
emilmont 1:fdd22bb7aa52 324
emilmont 1:fdd22bb7aa52 325 #ifndef ARM_MATH_BIG_ENDIAN
emilmont 1:fdd22bb7aa52 326 x3 = __PKHBT(x2, x1, 0);
emilmont 1:fdd22bb7aa52 327 #else
emilmont 1:fdd22bb7aa52 328 x3 = __PKHBT(x1, x2, 0);
emilmont 1:fdd22bb7aa52 329 #endif
emilmont 1:fdd22bb7aa52 330
emilmont 1:fdd22bb7aa52 331 acc3 = __SMLADX(x3, y1, acc3);
emilmont 1:fdd22bb7aa52 332
emilmont 1:fdd22bb7aa52 333 pScr2 += 4u;
emilmont 1:fdd22bb7aa52 334
emilmont 1:fdd22bb7aa52 335
emilmont 1:fdd22bb7aa52 336 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 337 tapCnt--;
emilmont 1:fdd22bb7aa52 338 }
emilmont 1:fdd22bb7aa52 339
emilmont 1:fdd22bb7aa52 340
emilmont 1:fdd22bb7aa52 341
emilmont 1:fdd22bb7aa52 342 /* Update scratch pointer for remaining samples of smaller length sequence */
emilmont 1:fdd22bb7aa52 343 pScr1 -= 4u;
emilmont 1:fdd22bb7aa52 344
emilmont 1:fdd22bb7aa52 345
emilmont 1:fdd22bb7aa52 346 /* apply same above for remaining samples of smaller length sequence */
emilmont 1:fdd22bb7aa52 347 tapCnt = (srcBLen) & 3u;
emilmont 1:fdd22bb7aa52 348
emilmont 1:fdd22bb7aa52 349 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 350 {
emilmont 1:fdd22bb7aa52 351
emilmont 1:fdd22bb7aa52 352 /* accumlate the results */
emilmont 1:fdd22bb7aa52 353 acc0 += (*pScr1++ * *pScr2);
emilmont 1:fdd22bb7aa52 354 acc1 += (*pScr1++ * *pScr2);
emilmont 1:fdd22bb7aa52 355 acc2 += (*pScr1++ * *pScr2);
emilmont 1:fdd22bb7aa52 356 acc3 += (*pScr1++ * *pScr2++);
emilmont 1:fdd22bb7aa52 357
emilmont 1:fdd22bb7aa52 358 pScr1 -= 3u;
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 blkCnt--;
emilmont 1:fdd22bb7aa52 365
emilmont 1:fdd22bb7aa52 366 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 367 out0 = (q7_t) (__SSAT(acc0 >> 7u, 8));
emilmont 1:fdd22bb7aa52 368 out1 = (q7_t) (__SSAT(acc1 >> 7u, 8));
emilmont 1:fdd22bb7aa52 369 out2 = (q7_t) (__SSAT(acc2 >> 7u, 8));
emilmont 1:fdd22bb7aa52 370 out3 = (q7_t) (__SSAT(acc3 >> 7u, 8));
emilmont 1:fdd22bb7aa52 371
emilmont 1:fdd22bb7aa52 372 *__SIMD32(pOut)++ = __PACKq7(out0, out1, out2, out3);
emilmont 1:fdd22bb7aa52 373
emilmont 1:fdd22bb7aa52 374 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 375 pScr2 = py;
emilmont 1:fdd22bb7aa52 376
emilmont 1:fdd22bb7aa52 377 pScratch1 += 4u;
emilmont 1:fdd22bb7aa52 378
emilmont 1:fdd22bb7aa52 379 }
emilmont 1:fdd22bb7aa52 380
emilmont 1:fdd22bb7aa52 381
emilmont 1:fdd22bb7aa52 382 blkCnt = (srcALen + srcBLen - 1u) & 0x3;
emilmont 1:fdd22bb7aa52 383
emilmont 1:fdd22bb7aa52 384 /* Calculate convolution for remaining samples of Bigger length sequence */
emilmont 1:fdd22bb7aa52 385 while(blkCnt > 0)
emilmont 1:fdd22bb7aa52 386 {
emilmont 1:fdd22bb7aa52 387 /* Initialze temporary scratch pointer as scratch1 */
emilmont 1:fdd22bb7aa52 388 pScr1 = pScratch1;
emilmont 1:fdd22bb7aa52 389
emilmont 1:fdd22bb7aa52 390 /* Clear Accumlators */
emilmont 1:fdd22bb7aa52 391 acc0 = 0;
emilmont 1:fdd22bb7aa52 392
emilmont 1:fdd22bb7aa52 393 tapCnt = (srcBLen) >> 1u;
emilmont 1:fdd22bb7aa52 394
emilmont 1:fdd22bb7aa52 395 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 396 {
emilmont 1:fdd22bb7aa52 397 acc0 += (*pScr1++ * *pScr2++);
emilmont 1:fdd22bb7aa52 398 acc0 += (*pScr1++ * *pScr2++);
emilmont 1:fdd22bb7aa52 399
emilmont 1:fdd22bb7aa52 400 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 401 tapCnt--;
emilmont 1:fdd22bb7aa52 402 }
emilmont 1:fdd22bb7aa52 403
emilmont 1:fdd22bb7aa52 404 tapCnt = (srcBLen) & 1u;
emilmont 1:fdd22bb7aa52 405
emilmont 1:fdd22bb7aa52 406 /* apply same above for remaining samples of smaller length sequence */
emilmont 1:fdd22bb7aa52 407 while(tapCnt > 0u)
emilmont 1:fdd22bb7aa52 408 {
emilmont 1:fdd22bb7aa52 409
emilmont 1:fdd22bb7aa52 410 /* accumlate the results */
emilmont 1:fdd22bb7aa52 411 acc0 += (*pScr1++ * *pScr2++);
emilmont 1:fdd22bb7aa52 412
emilmont 1:fdd22bb7aa52 413 /* Decrement the loop counter */
emilmont 1:fdd22bb7aa52 414 tapCnt--;
emilmont 1:fdd22bb7aa52 415 }
emilmont 1:fdd22bb7aa52 416
emilmont 1:fdd22bb7aa52 417 blkCnt--;
emilmont 1:fdd22bb7aa52 418
emilmont 1:fdd22bb7aa52 419 /* Store the result in the accumulator in the destination buffer. */
emilmont 1:fdd22bb7aa52 420 *pOut++ = (q7_t) (__SSAT(acc0 >> 7u, 8));
emilmont 1:fdd22bb7aa52 421
emilmont 1:fdd22bb7aa52 422 /* Initialization of inputB pointer */
emilmont 1:fdd22bb7aa52 423 pScr2 = py;
emilmont 1:fdd22bb7aa52 424
emilmont 1:fdd22bb7aa52 425 pScratch1 += 1u;
emilmont 1:fdd22bb7aa52 426
emilmont 1:fdd22bb7aa52 427 }
emilmont 1:fdd22bb7aa52 428
emilmont 1:fdd22bb7aa52 429 }
emilmont 1:fdd22bb7aa52 430
emilmont 1:fdd22bb7aa52 431
emilmont 1:fdd22bb7aa52 432 /**
emilmont 1:fdd22bb7aa52 433 * @} end of Conv group
emilmont 1:fdd22bb7aa52 434 */