arm_mult_q7.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_mult_q7.c 00009 * 00010 * Description: Q7 vector multiplication. 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 00053 void arm_mult_q7( 00054 q7_t * pSrcA, 00055 q7_t * pSrcB, 00056 q7_t * pDst, 00057 uint32_t blockSize) 00058 { 00059 uint32_t blkCnt; /* loop counters */ 00060 q7_t inA1, inA2, inA3, inA4; /* temporary input variables */ 00061 q7_t inB1, inB2, inB3, inB4; /* temporary input variables */ 00062 q15_t out1, out2, out3, out4; /* temporary output variables */ 00063 00064 /* loop Unrolling */ 00065 blkCnt = blockSize >> 3u; 00066 00067 /* First part of the processing with loop unrolling. Compute 8 outputs at a time. 00068 ** a second loop below computes the remaining 1 to 7 samples. */ 00069 while(blkCnt > 0u) 00070 { 00071 /* C = A * B */ 00072 /* Multiply the inputs and store the results in temporary variables */ 00073 /* read sample from sourceA */ 00074 inA1 = *pSrcA; 00075 /* read sample from sourceB */ 00076 inB1 = *pSrcB; 00077 /* read sample from sourceA */ 00078 inA2 = *(pSrcA + 1); 00079 /* read sample from sourceB */ 00080 inB2 = *(pSrcB + 1); 00081 00082 /* out = sourceA * sourceB output in q15_t format */ 00083 out1 = (q15_t)((q7_t) (inA1) * (inB1)); 00084 00085 /* read sample from sourceA */ 00086 inA3 = *(pSrcA + 2); 00087 /* read sample from sourceB */ 00088 inB3 = *(pSrcB + 2); 00089 00090 /* out = sourceA * sourceB output in q15_t format */ 00091 out2 = (q15_t)((q7_t) (inA2) * (inB2)); 00092 00093 /* read sample from sourceA */ 00094 inA4 = *(pSrcA + 3); 00095 00096 /* convert output to 1.7 format */ 00097 out1 = out1 >> 7u; 00098 00099 /* read sample from sourceB */ 00100 inB4 = *(pSrcB + 3); 00101 00102 /* out = sourceA * sourceB output in q15_t format */ 00103 out3 = (q15_t)((q7_t) (inA3) * (inB3)); 00104 out4 = (q15_t)((q7_t) (inA4) * (inB4)); 00105 00106 /* convert output to 1.7 format */ 00107 out2 = out2 >> 7u; 00108 00109 /* read sample from sourceA */ 00110 inA1 = *(pSrcA + 4); 00111 00112 /* convert output to 1.7 format */ 00113 out3 = out3 >> 7u; 00114 00115 /* read sample from sourceB */ 00116 inB1 = *(pSrcB + 4); 00117 00118 /* convert output to 1.7 format */ 00119 out4 = out4 >> 7u; 00120 00121 /* saturate the result */ 00122 #ifdef CCS 00123 00124 inA3 = (q7_t) __SSATA(out1, 0, 8); 00125 inA4 = (q7_t) __SSATA(out2, 0, 8); 00126 inB3 = (q7_t) __SSATA(out3, 0, 8); 00127 inB4 = (q7_t) __SSATA(out4, 0, 8); 00128 00129 #else 00130 00131 inA3 = (q7_t) __SSAT(out1, 8); 00132 inA4 = (q7_t) __SSAT(out2, 8); 00133 inB3 = (q7_t) __SSAT(out3, 8); 00134 inB4 = (q7_t) __SSAT(out4, 8); 00135 00136 #endif // #ifdef CCS 00137 00138 /* Store the results of 4 inputs in the destination buffer in single cycle by packing */ 00139 *__SIMD32(pDst)++ = __PACKq7(inA3, inA4, inB3, inB4); 00140 00141 /* read sample from sourceA */ 00142 inA2 = *(pSrcA + 5); 00143 /* read sample from sourceB */ 00144 inB2 = *(pSrcB + 5); 00145 00146 /* out = sourceA * sourceB output in q15_t format */ 00147 out1 = (q15_t)((q7_t) (inA1) * (inB1)); 00148 00149 /* read sample from sourceA */ 00150 inA3 = *(pSrcA + 6); 00151 /* read sample from sourceB */ 00152 inB3 = *(pSrcB + 6); 00153 00154 /* out = sourceA * sourceB output in q15_t format */ 00155 out2 = (q15_t)((q7_t) (inA2) * (inB2)); 00156 00157 /* read sample from sourceA */ 00158 inA4 = *(pSrcA + 7); 00159 00160 /* convert output to 1.7 format */ 00161 out1 = out1 >> 7u; 00162 00163 /* read sample from sourceB */ 00164 inB4 = *(pSrcB + 7); 00165 00166 /* out = sourceA * sourceB output in q15_t format */ 00167 out3 = (q15_t)((q7_t) (inA3) * (inB3)); 00168 out4 = (q15_t)((q7_t) (inA4) * (inB4)); 00169 00170 /* convert output to 1.7 format */ 00171 out2 = out2 >> 7u; 00172 out3 = out3 >> 7u; 00173 out4 = out4 >> 7u; 00174 00175 /* saturate the result */ 00176 #ifdef CCS 00177 00178 inA1 = (q7_t) __SSATA(out1, 0, 8); 00179 inA2 = (q7_t) __SSATA(out2, 0, 8); 00180 inA3 = (q7_t) __SSATA(out3, 0, 8); 00181 inA4 = (q7_t) __SSATA(out4, 0, 8); 00182 00183 #else 00184 00185 inA1 = (q7_t) __SSAT(out1, 8); 00186 inA2 = (q7_t) __SSAT(out2, 8); 00187 inA3 = (q7_t) __SSAT(out3, 8); 00188 inA4 = (q7_t) __SSAT(out4, 8); 00189 00190 #endif // #ifdef CCS 00191 00192 /* Store the results of 4 inputs in the destination buffer in single cycle by packing */ 00193 *__SIMD32(pDst)++ = __PACKq7(inA1, inA2, inA3, inA4); 00194 00195 /* increment sourceA by 8 to process next samples */ 00196 pSrcA += 8u; 00197 /* increment sourceB by 8 to process next samples */ 00198 pSrcB += 8u; 00199 00200 /* Decrement the blockSize loop counter */ 00201 blkCnt--; 00202 } 00203 00204 /* If the blockSize is not a multiple of 8, compute any remaining output samples here. 00205 ** No loop unrolling is used. */ 00206 blkCnt = blockSize % 0x8u; 00207 00208 while(blkCnt > 0u) 00209 { 00210 /* C = A * B */ 00211 /* Multiply the inputs and store the result in the destination buffer */ 00212 *pDst++ = (q7_t) (((q15_t) (*pSrcA++) * (*pSrcB++)) >> 7); 00213 00214 /* Decrement the blockSize loop counter */ 00215 blkCnt--; 00216 } 00217 } 00218
