arm_fir_lattice_f32.c
Go to the documentation of this file.
00001 /* ---------------------------------------------------------------------- 00002 * Copyright (C) 2011 ARM Limited. All rights reserved. 00003 * 00004 * $Date: 15. December 2011 00005 * $Revision: V2.0.0 00006 * 00007 * Project: Cortex-R DSP Library 00008 * Title: arm_fir_lattice_f32.c 00009 * 00010 * Description: Processing function for the floating-point FIR Lattice filter. 00011 * 00012 * Target Processor: Cortex-R4/R5 00013 * 00014 * Version 1.0.0 2011/03/08 00015 * Alpha release. 00016 * 00017 * Version 1.0.1 2011/09/30 00018 * Beta release. 00019 * 00020 * Version 2.0.0 2011/12/15 00021 * Final release. 00022 * 00023 * -------------------------------------------------------------------- */ 00024 #include "arm_math.h" 00025 00112 void arm_fir_lattice_f32( 00113 const arm_fir_lattice_instance_f32 * S, 00114 float32_t * pSrc, 00115 float32_t * pDst, 00116 uint32_t blockSize) 00117 { 00118 float32_t *pState; /* State pointer */ 00119 float32_t *pCoeffs = S->pCoeffs; /* Coefficient pointer */ 00120 float32_t *px; /* temporary state pointer */ 00121 float32_t *pk, k; /* temporary coefficient pointer */ 00122 float32_t fcurr1, fnext1, gcurr1, gnext1; /* temporary variables for first sample in loop unrolling */ 00123 float32_t fcurr2, fnext2, gnext2; /* temporary variables for second sample in loop unrolling */ 00124 float32_t fcurr3, fnext3, gnext3; /* temporary variables for third sample in loop unrolling */ 00125 float32_t fcurr4, fnext4, gnext4; /* temporary variables for fourth sample in loop unrolling */ 00126 uint32_t numStages = S->numStages; /* Number of stages in the filter */ 00127 uint32_t blkCnt, stageCnt; /* temporary variables for counts */ 00128 00129 pState = &S->pState[0]; 00130 00131 blkCnt = blockSize >> 2; 00132 00133 /* First part of the processing with loop unrolling. Compute 4 outputs at a time. 00134 a second loop below computes the remaining 1 to 3 samples. */ 00135 while(blkCnt > 0u) 00136 { 00137 00138 /* Read two samples from input buffer */ 00139 /* f0(n) = x(n) */ 00140 fcurr1 = *pSrc++; 00141 fcurr2 = *pSrc++; 00142 00143 /* Initialize coeff pointer */ 00144 pk = (pCoeffs); 00145 00146 /* Initialize state pointer */ 00147 px = pState; 00148 00149 k = *pk++; 00150 00151 /* Read g0(n-1) from state */ 00152 gcurr1 = *px; 00153 00154 /* Process first sample for first tap */ 00155 /* f1(n) = f0(n) + K1 * g0(n-1) */ 00156 fnext1 = fcurr1 + (k * gcurr1); 00157 00158 /* g1(n) = f0(n) * K1 + g0(n-1) */ 00159 gnext1 = (fcurr1 * k) + gcurr1; 00160 00161 /* Process second sample for first tap */ 00162 /* for sample 2 processing */ 00163 fnext2 = fcurr2 + (k * fcurr1); 00164 gnext2 = (fcurr2 * k) + fcurr1; 00165 00166 /* Read next two samples from input buffer */ 00167 /* f0(n+2) = x(n+2) */ 00168 fcurr3 = *pSrc++; 00169 fcurr4 = *pSrc++; 00170 00171 /* Copy only last input samples into the state buffer 00172 which will be used for next four samples processing */ 00173 *px++ = fcurr4; 00174 00175 /* Process third sample for first tap */ 00176 fnext3 = fcurr3 + (k * fcurr2); 00177 gnext3 = (fcurr3 * k) + fcurr2; 00178 00179 /* Process fourth sample for first tap */ 00180 fnext4 = fcurr4 + (k * fcurr3); 00181 gnext4 = (fcurr4 * k) + fcurr3; 00182 00183 /* Update of f values for next coefficient set processing */ 00184 fcurr1 = fnext1; 00185 fcurr2 = fnext2; 00186 fcurr3 = fnext3; 00187 fcurr4 = fnext4; 00188 00189 /* Loop unrolling. Process 4 taps at a time . */ 00190 stageCnt = (numStages - 1u) >> 2u; 00191 00192 /* Loop over the number of taps. Unroll by a factor of 4. 00193 ** Repeat until we've computed numStages-3 coefficients. */ 00194 00195 /* Process 2nd, 3rd, 4th and 5th taps ... here */ 00196 while(stageCnt > 0u) 00197 { 00198 /* Read g1(n-1), g3(n-1) .... from state */ 00199 gcurr1 = *px; 00200 00201 /* Read the reflection coefficient */ 00202 k = *pk; 00203 00204 /* save g1(n) in state buffer */ 00205 *px++ = gnext4; 00206 00207 /* Process first sample for 2nd, 6th .. tap */ 00208 /* Sample processing for K2, K6.... */ 00209 /* f2(n) = f1(n) + K2 * g1(n-1) */ 00210 fnext1 = fcurr1 + (k * gcurr1); 00211 /* Process second sample for 2nd, 6th .. tap */ 00212 /* for sample 2 processing */ 00213 fnext2 = fcurr2 + (k * gnext1); 00214 /* Process third sample for 2nd, 6th .. tap */ 00215 fnext3 = fcurr3 + (k * gnext2); 00216 /* Process fourth sample for 2nd, 6th .. tap */ 00217 fnext4 = fcurr4 + (k * gnext3); 00218 00219 /* g2(n) = f1(n) * K2 + g1(n-1) */ 00220 /* Calculation of state values for next stage */ 00221 gnext4 = (fcurr4 * k) + gnext3; 00222 gnext3 = (fcurr3 * k) + gnext2; 00223 gnext2 = (fcurr2 * k) + gnext1; 00224 gnext1 = (fcurr1 * k) + gcurr1; 00225 00226 00227 /* Read g2(n-1), g4(n-1) .... from state */ 00228 gcurr1 = *px; 00229 00230 /* Read the reflection coefficient */ 00231 k = *(pk + 1u); 00232 00233 /* save g2(n) in state buffer */ 00234 *px++ = gnext4; 00235 00236 /* Sample processing for K3, K7.... */ 00237 /* Process first sample for 3rd, 7th .. tap */ 00238 /* f3(n) = f2(n) + K3 * g2(n-1) */ 00239 fcurr1 = fnext1 + (k * gcurr1); 00240 /* Process second sample for 3rd, 7th .. tap */ 00241 fcurr2 = fnext2 + (k * gnext1); 00242 /* Process third sample for 3rd, 7th .. tap */ 00243 fcurr3 = fnext3 + (k * gnext2); 00244 /* Process fourth sample for 3rd, 7th .. tap */ 00245 fcurr4 = fnext4 + (k * gnext3); 00246 00247 /* Calculation of state values for next stage */ 00248 /* g3(n) = f2(n) * K3 + g2(n-1) */ 00249 gnext4 = (fnext4 * k) + gnext3; 00250 gnext3 = (fnext3 * k) + gnext2; 00251 gnext2 = (fnext2 * k) + gnext1; 00252 gnext1 = (fnext1 * k) + gcurr1; 00253 00254 00255 /* Read g1(n-1), g3(n-1) .... from state */ 00256 gcurr1 = *px; 00257 00258 /* Read the reflection coefficient */ 00259 k = *(pk + 2u); 00260 00261 /* save g3(n) in state buffer */ 00262 *px++ = gnext4; 00263 00264 /* Sample processing for K4, K8.... */ 00265 /* Process first sample for 4th, 8th .. tap */ 00266 /* f4(n) = f3(n) + K4 * g3(n-1) */ 00267 fnext1 = fcurr1 + (k * gcurr1); 00268 /* Process second sample for 4th, 8th .. tap */ 00269 /* for sample 2 processing */ 00270 fnext2 = fcurr2 + (k * gnext1); 00271 /* Process third sample for 4th, 8th .. tap */ 00272 fnext3 = fcurr3 + (k * gnext2); 00273 /* Process fourth sample for 4th, 8th .. tap */ 00274 fnext4 = fcurr4 + (k * gnext3); 00275 00276 /* g4(n) = f3(n) * K4 + g3(n-1) */ 00277 /* Calculation of state values for next stage */ 00278 gnext4 = (fcurr4 * k) + gnext3; 00279 gnext3 = (fcurr3 * k) + gnext2; 00280 gnext2 = (fcurr2 * k) + gnext1; 00281 gnext1 = (fcurr1 * k) + gcurr1; 00282 00283 /* Read g2(n-1), g4(n-1) .... from state */ 00284 gcurr1 = *px; 00285 00286 /* Read the reflection coefficient */ 00287 k = *(pk + 3u); 00288 00289 /* save g4(n) in state buffer */ 00290 *px++ = gnext4; 00291 00292 /* update Coeff pointer */ 00293 pk += 4u; 00294 00295 /* Sample processing for K5, K9.... */ 00296 /* Process first sample for 5th, 9th .. tap */ 00297 /* f5(n) = f4(n) + K5 * g4(n-1) */ 00298 fcurr1 = fnext1 + (k * gcurr1); 00299 /* Process second sample for 5th, 9th .. tap */ 00300 fcurr2 = fnext2 + (k * gnext1); 00301 /* Process third sample for 5th, 9th .. tap */ 00302 fcurr3 = fnext3 + (k * gnext2); 00303 /* Process fourth sample for 5th, 9th .. tap */ 00304 fcurr4 = fnext4 + (k * gnext3); 00305 00306 /* Calculation of state values for next stage */ 00307 /* g5(n) = f4(n) * K5 + g4(n-1) */ 00308 gnext4 = (fnext4 * k) + gnext3; 00309 gnext3 = (fnext3 * k) + gnext2; 00310 gnext2 = (fnext2 * k) + gnext1; 00311 gnext1 = (fnext1 * k) + gcurr1; 00312 00313 00314 00315 stageCnt--; 00316 } 00317 00318 /* If the (filter length -1) is not a multiple of 4, compute the remaining filter taps */ 00319 stageCnt = (numStages - 1u) % 0x4u; 00320 00321 while(stageCnt > 0u) 00322 { 00323 gcurr1 = *px; 00324 00325 /* save g value in state buffer */ 00326 *px++ = gnext4; 00327 00328 /* Read the reflection coefficient */ 00329 k = (*pk++); 00330 00331 /* Process four samples for last three taps here */ 00332 fnext1 = fcurr1 + (k * gcurr1); 00333 fnext2 = fcurr2 + (k * gnext1); 00334 fnext3 = fcurr3 + (k * gnext2); 00335 fnext4 = fcurr4 + (k * gnext3); 00336 00337 /* g1(n) = f0(n) * K1 + g0(n-1) */ 00338 gnext4 = (fcurr4 * k) + gnext3; 00339 gnext3 = (fcurr3 * k) + gnext2; 00340 gnext2 = (fcurr2 * k) + gnext1; 00341 gnext1 = (fcurr1 * k) + gcurr1; 00342 00343 /* Update of f values for next coefficient set processing */ 00344 fcurr1 = fnext1; 00345 fcurr2 = fnext2; 00346 fcurr3 = fnext3; 00347 fcurr4 = fnext4; 00348 00349 stageCnt--; 00350 00351 } 00352 00353 /* The results in the 4 accumulators, store in the destination buffer. */ 00354 /* y(n) = fN(n) */ 00355 *pDst++ = fcurr1; 00356 *pDst++ = fcurr2; 00357 *pDst++ = fcurr3; 00358 *pDst++ = fcurr4; 00359 00360 blkCnt--; 00361 } 00362 00363 /* If the blockSize is not a multiple of 4, compute any remaining output samples here. 00364 ** No loop unrolling is used. */ 00365 blkCnt = blockSize % 0x4u; 00366 00367 while(blkCnt > 0u) 00368 { 00369 /* f0(n) = x(n) */ 00370 fcurr1 = *pSrc++; 00371 00372 /* Initialize coeff pointer */ 00373 pk = (pCoeffs); 00374 00375 /* Initialize state pointer */ 00376 px = pState; 00377 00378 /* read g2(n) from state buffer */ 00379 gcurr1 = *px; 00380 00381 /* Read the reflection coefficient */ 00382 k = (*pk++); 00383 00384 /* for sample 1 processing */ 00385 /* f1(n) = f0(n) + K1 * g0(n-1) */ 00386 fnext1 = fcurr1 + (k * gcurr1); 00387 /* g1(n) = f0(n) * K1 + g0(n-1) */ 00388 gnext1 = (fcurr1 * k) + gcurr1; 00389 00390 /* save g1(n) in state buffer */ 00391 *px++ = fcurr1; 00392 00393 /* f1(n) is saved in fcurr1 00394 for next stage processing */ 00395 fcurr1 = fnext1; 00396 00397 stageCnt = (numStages - 1u); 00398 00399 /* stage loop */ 00400 while(stageCnt > 0u) 00401 { 00402 /* read g2(n) from state buffer */ 00403 gcurr1 = *px; 00404 00405 /* Read the reflection coefficient */ 00406 k = (*pk++); 00407 00408 /* save g1(n) in state buffer */ 00409 *px++ = gnext1; 00410 00411 /* Sample processing for K2, K3.... */ 00412 /* f2(n) = f1(n) + K2 * g1(n-1) */ 00413 fnext1 = fcurr1 + (k * gcurr1); 00414 /* g2(n) = f1(n) * K2 + g1(n-1) */ 00415 gnext1 = (fcurr1 * k) + gcurr1; 00416 00417 /* f1(n) is saved in fcurr1 00418 for next stage processing */ 00419 fcurr1 = fnext1; 00420 00421 stageCnt--; 00422 00423 } 00424 00425 /* y(n) = fN(n) */ 00426 *pDst++ = fcurr1; 00427 00428 blkCnt--; 00429 00430 } 00431 } 00432
