arm_sub_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_sub_q7.c 00009 * 00010 * Description: Q7 vector subtraction. 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_sub_q7( 00054 q7_t * pSrcA, 00055 q7_t * pSrcB, 00056 q7_t * pDst, 00057 uint32_t blockSize) 00058 { 00059 uint32_t blkCnt; /* loop counter */ 00060 q31_t inA1, inB1, inA2, inB2; /* temporary input variabels */ 00061 q7_t inA, inB; /* temporary variables */ 00062 q31_t out1, out2, out3, out4; /* temporary output variables */ 00063 00064 00065 /*loop Unrolling */ 00066 blkCnt = blockSize >> 4u; 00067 00068 /* First part of the processing with loop unrolling. Compute 16 outputs at a time. 00069 ** a second loop below computes the remaining 1 to 15 samples. */ 00070 while(blkCnt > 0u) 00071 { 00072 /* C = A - B */ 00073 /* Subtract and then store the results in the destination buffer 4 samples at a time. */ 00074 /* read 4 samples at a time from sourceA */ 00075 inA1 = _SIMD32_OFFSET(pSrcA); 00076 /* read 4 samples at a time from sourceB */ 00077 inB1 = _SIMD32_OFFSET(pSrcB); 00078 /* read 4 samples at a time from sourceA */ 00079 inA2 = _SIMD32_OFFSET(pSrcA + 4); 00080 00081 /* out = saturate(sourceA - sourceB) four samples at a time */ 00082 out1 = __QSUB8(inA1, inB1); 00083 00084 /* read 4 samples at a time from sourceB */ 00085 inB2 = _SIMD32_OFFSET(pSrcB + 4); 00086 00087 /* store result to destination four samples at a time */ 00088 _SIMD32_OFFSET(pDst) = out1; 00089 00090 /* out = saturate(sourceA - sourceB) four samples at a time */ 00091 out2 = __QSUB8(inA2, inB2); 00092 00093 /* read 4 samples at a time from sourceA */ 00094 inA1 = _SIMD32_OFFSET(pSrcA + 8); 00095 /* read 4 samples at a time from sourceB */ 00096 inB1 = _SIMD32_OFFSET(pSrcB + 8); 00097 /* read 4 samples at a time from sourceA */ 00098 inA2 = _SIMD32_OFFSET(pSrcA + 12); 00099 00100 /* out = saturate(sourceA - sourceB) four samples at a time */ 00101 out3 = __QSUB8(inA1, inB1); 00102 00103 /* read 4 samples at a time from sourceB */ 00104 inB2 = _SIMD32_OFFSET(pSrcB + 12); 00105 00106 /* increment sourceA pointer by 16 to process next samples */ 00107 pSrcA += 16u; 00108 00109 /* store result to destination four samples at a time */ 00110 _SIMD32_OFFSET(pDst + 4) = out2; 00111 00112 /* out = saturate(sourceA - sourceB) four samples at a time */ 00113 out4 = __QSUB8(inA2, inB2); 00114 00115 /* store result to destination four samples at a time */ 00116 _SIMD32_OFFSET(pDst + 8) = out3; 00117 00118 /* Update source pointer to process next sampels */ 00119 pSrcB += 16u; 00120 00121 /* store result to destination four samples at a time */ 00122 _SIMD32_OFFSET(pDst + 12) = out4; 00123 00124 /* Update destination pointer to process next sampels */ 00125 pDst += 16u; 00126 00127 /* Decrement the loop counter */ 00128 blkCnt--; 00129 } 00130 00131 /* If the blockSize is not a multiple of 16, compute any remaining output samples here. 00132 ** No loop unrolling is used. */ 00133 blkCnt = blockSize % 0x10u; 00134 00135 while(blkCnt > 0u) 00136 { 00137 /* C = A - B */ 00138 /* Subtract and then store the result in the destination buffer. */ 00139 inA = *pSrcA++; 00140 inB = *pSrcB++; 00141 #ifdef CCS 00142 *pDst++ = __SSATA(inA - inB, 0, 8); 00143 #else 00144 *pDst++ = __SSAT(inA - inB, 8); 00145 #endif //#ifdef CCS 00146 00147 /* Decrement the loop counter */ 00148 blkCnt--; 00149 } 00150 00151 } 00152
