arm_biquad_cascade_df1_fast_q31.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_biquad_cascade_df1_fast_q31.c 00009 * 00010 * Description: Processing function for the 00011 * Q31 Fast Biquad cascade DirectFormI(DF1) filter. 00012 * 00013 * Target Processor: Cortex-R4/R5 00014 * 00015 * Version 1.0.0 2011/03/08 00016 * Alpha release. 00017 * 00018 * Version 1.0.1 2011/09/30 00019 * Beta release. 00020 * 00021 * Version 2.0.0 2011/12/15 00022 * Final release. 00023 * 00024 * -------------------------------------------------------------------- */ 00025 00026 #include "arm_math.h" 00027 00061 void arm_biquad_cascade_df1_fast_q31( 00062 const arm_biquad_casd_df1_inst_q31 * S, 00063 q31_t * pSrc, 00064 q31_t * pDst, 00065 uint32_t blockSize) 00066 { 00067 q31_t acc; /* accumulator */ 00068 q31_t Xn1, Xn2, Yn1, Yn2; /* Filter state variables */ 00069 q31_t b0, b1, b2, a1, a2; /* Filter coefficients */ 00070 q31_t *pIn = pSrc; /* input pointer initialization */ 00071 q31_t *pOut = pDst; /* output pointer initialization */ 00072 q31_t *pState = S->pState; /* pState pointer initialization */ 00073 q31_t *pCoeffs = S->pCoeffs; /* coeff pointer initialization */ 00074 q31_t Xn; /* temporary input */ 00075 int32_t shift = (int32_t) S->postShift + 1; /* Shift to be applied to the output */ 00076 uint32_t sample, stage = S->numStages; /* loop counters */ 00077 00078 00079 do 00080 { 00081 /* Reading the coefficients */ 00082 b0 = *pCoeffs++; 00083 b1 = *pCoeffs++; 00084 b2 = *pCoeffs++; 00085 a1 = *pCoeffs++; 00086 a2 = *pCoeffs++; 00087 00088 /* Reading the state values */ 00089 Xn1 = pState[0]; 00090 Xn2 = pState[1]; 00091 Yn1 = pState[2]; 00092 Yn2 = pState[3]; 00093 00094 /* Apply loop unrolling and compute 4 output values simultaneously. */ 00095 /* The variables acc ... acc3 hold output values that are being computed: 00096 * 00097 * acc = b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2] 00098 */ 00099 00100 sample = blockSize >> 2u; 00101 00102 /* First part of the processing with loop unrolling. Compute 4 outputs at a time. 00103 ** a second loop below computes the remaining 1 to 3 samples. */ 00104 while(sample > 0u) 00105 { 00106 /* Read the input */ 00107 Xn = *pIn; 00108 00109 /* acc = b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2] */ 00110 /* acc = b0 * x[n] */ 00111 acc = (q31_t) (((q63_t) b1 * Xn1) >> 32); 00112 /* acc += b1 * x[n-1] */ 00113 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b0 * (Xn))) >> 32); 00114 /* acc += b[2] * x[n-2] */ 00115 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b2 * (Xn2))) >> 32); 00116 /* acc += a1 * y[n-1] */ 00117 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a1 * (Yn1))) >> 32); 00118 /* acc += a2 * y[n-2] */ 00119 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a2 * (Yn2))) >> 32); 00120 00121 /* The result is converted to 1.31 , Yn2 variable is reused */ 00122 Yn2 = acc << shift; 00123 00124 /* Read the second input */ 00125 Xn2 = *(pIn + 1u); 00126 00127 /* Store the output in the destination buffer. */ 00128 *pOut = Yn2; 00129 00130 /* acc = b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2] */ 00131 /* acc = b0 * x[n] */ 00132 acc = (q31_t) (((q63_t) b0 * (Xn2)) >> 32); 00133 /* acc += b1 * x[n-1] */ 00134 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b1 * (Xn))) >> 32); 00135 /* acc += b[2] * x[n-2] */ 00136 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b2 * (Xn1))) >> 32); 00137 /* acc += a1 * y[n-1] */ 00138 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a1 * (Yn2))) >> 32); 00139 /* acc += a2 * y[n-2] */ 00140 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a2 * (Yn1))) >> 32); 00141 00142 /* The result is converted to 1.31, Yn1 variable is reused */ 00143 Yn1 = acc << shift; 00144 00145 /* Read the third input */ 00146 Xn1 = *(pIn + 2u); 00147 00148 /* Store the output in the destination buffer. */ 00149 *(pOut + 1u) = Yn1; 00150 00151 /* acc = b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2] */ 00152 /* acc = b0 * x[n] */ 00153 acc = (q31_t) (((q63_t) b0 * (Xn1)) >> 32); 00154 /* acc += b1 * x[n-1] */ 00155 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b1 * (Xn2))) >> 32); 00156 /* acc += b[2] * x[n-2] */ 00157 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b2 * (Xn))) >> 32); 00158 /* acc += a1 * y[n-1] */ 00159 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a1 * (Yn1))) >> 32); 00160 /* acc += a2 * y[n-2] */ 00161 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a2 * (Yn2))) >> 32); 00162 00163 /* The result is converted to 1.31, Yn2 variable is reused */ 00164 Yn2 = acc << shift; 00165 00166 /* Read the forth input */ 00167 Xn = *(pIn + 3u); 00168 00169 /* Store the output in the destination buffer. */ 00170 *(pOut + 2u) = Yn2; 00171 pIn += 4u; 00172 00173 /* acc = b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2] */ 00174 /* acc = b0 * x[n] */ 00175 acc = (q31_t) (((q63_t) b0 * (Xn)) >> 32); 00176 /* acc += b1 * x[n-1] */ 00177 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b1 * (Xn1))) >> 32); 00178 /* acc += b[2] * x[n-2] */ 00179 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b2 * (Xn2))) >> 32); 00180 /* acc += a1 * y[n-1] */ 00181 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a1 * (Yn2))) >> 32); 00182 /* acc += a2 * y[n-2] */ 00183 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a2 * (Yn1))) >> 32); 00184 00185 /* Every time after the output is computed state should be updated. */ 00186 /* The states should be updated as: */ 00187 /* Xn2 = Xn1 */ 00188 Xn2 = Xn1; 00189 00190 /* The result is converted to 1.31, Yn1 variable is reused */ 00191 Yn1 = acc << shift; 00192 00193 /* Xn1 = Xn */ 00194 Xn1 = Xn; 00195 00196 /* Store the output in the destination buffer. */ 00197 *(pOut + 3u) = Yn1; 00198 pOut += 4u; 00199 00200 /* decrement the loop counter */ 00201 sample--; 00202 } 00203 00204 /* If the blockSize is not a multiple of 4, compute any remaining output samples here. 00205 ** No loop unrolling is used. */ 00206 sample = (blockSize & 0x3u); 00207 00208 while(sample > 0u) 00209 { 00210 /* Read the input */ 00211 Xn = *pIn++; 00212 00213 /* acc = b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2] */ 00214 /* acc = b0 * x[n] */ 00215 acc = (q31_t) (((q63_t) b0 * (Xn)) >> 32); 00216 /* acc += b1 * x[n-1] */ 00217 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b1 * (Xn1))) >> 32); 00218 /* acc += b[2] * x[n-2] */ 00219 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) b2 * (Xn2))) >> 32); 00220 /* acc += a1 * y[n-1] */ 00221 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a1 * (Yn1))) >> 32); 00222 /* acc += a2 * y[n-2] */ 00223 acc = (q31_t) ((((q63_t) acc << 32) + ((q63_t) a2 * (Yn2))) >> 32); 00224 /* The result is converted to 1.31 */ 00225 acc = acc << shift; 00226 00227 /* Every time after the output is computed state should be updated. */ 00228 /* The states should be updated as: */ 00229 /* Xn2 = Xn1 */ 00230 /* Xn1 = Xn */ 00231 /* Yn2 = Yn1 */ 00232 /* Yn1 = acc */ 00233 Xn2 = Xn1; 00234 Xn1 = Xn; 00235 Yn2 = Yn1; 00236 Yn1 = acc; 00237 00238 /* Store the output in the destination buffer. */ 00239 *pOut++ = acc; 00240 00241 /* decrement the loop counter */ 00242 sample--; 00243 } 00244 00245 /* The first stage goes from the input buffer to the output buffer. */ 00246 /* Subsequent stages occur in-place in the output buffer */ 00247 pIn = pDst; 00248 00249 /* Reset to destination pointer */ 00250 pOut = pDst; 00251 00252 /* Store the updated state variables back into the pState array */ 00253 *pState++ = Xn1; 00254 *pState++ = Xn2; 00255 *pState++ = Yn1; 00256 *pState++ = Yn2; 00257 00258 } while(--stage); 00259 } 00260
