arm_mat_mult_q31.c

00001 /* ----------------------------------------------------------------------  
00002 * Copyright (C) 2010 ARM Limited. All rights reserved.  
00003 *  
00004 * $Date:        29. November 2010  
00005 * $Revision:    V1.0.3  
00006 *  
00007 * Project:      CMSIS DSP Library  
00008 * Title:        arm_mat_mult_q31.c  
00009 *  
00010 * Description:   Q31 matrix multiplication.  
00011 *  
00012 * Target Processor: Cortex-M4/Cortex-M3
00013 *  
00014 * Version 1.0.3 2010/11/29 
00015 *    Re-organized the CMSIS folders and updated documentation.  
00016 *   
00017 * Version 1.0.2 2010/11/11  
00018 *    Documentation updated.   
00019 *  
00020 * Version 1.0.1 2010/10/05   
00021 *    Production release and review comments incorporated.  
00022 *  
00023 * Version 1.0.0 2010/09/20   
00024 *    Production release and review comments incorporated.  
00025 *  
00026 * Version 0.0.5  2010/04/26   
00027 *    incorporated review comments and updated with latest CMSIS layer  
00028 *  
00029 * Version 0.0.3  2010/03/10   
00030 *    Initial version  
00031 * -------------------------------------------------------------------- */ 
00032  
00033 #include "arm_math.h" 
00034  
00035 /**  
00036  * @ingroup groupMatrix  
00037  */ 
00038  
00039 /**  
00040  * @addtogroup MatrixMult  
00041  * @{  
00042  */ 
00043  
00044 /**  
00045  * @brief Q31 matrix multiplication  
00046  * @param[in]       *pSrcA points to the first input matrix structure  
00047  * @param[in]       *pSrcB points to the second input matrix structure  
00048  * @param[out]      *pDst points to output matrix structure  
00049  * @return          The function returns either  
00050  * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.  
00051  *  
00052  * @details  
00053  * <b>Scaling and Overflow Behavior:</b>  
00054  *  
00055  * \par  
00056  * The function is implemented using an internal 64-bit accumulator.  
00057  * The accumulator has a 2.62 format and maintains full precision of the intermediate  
00058  * multiplication results but provides only a single guard bit. There is no saturation  
00059  * on intermediate additions. Thus, if the accumulator overflows it wraps around and  
00060  * distorts the result. The input signals should be scaled down to avoid intermediate  
00061  * overflows. The input is thus scaled down by log2(numColsA) bits  
00062  * to avoid overflows, as a total of numColsA additions are performed internally.  
00063  * The 2.62 accumulator is right shifted by 31 bits and saturated to 1.31 format to yield the final result.  
00064  *  
00065  * \par  
00066  * See <code>arm_mat_mult_fast_q31()</code> for a faster but less precise implementation of this function.  
00067  *  
00068  */ 
00069  
00070 arm_status arm_mat_mult_q31( 
00071   const arm_matrix_instance_q31 * pSrcA, 
00072   const arm_matrix_instance_q31 * pSrcB, 
00073   arm_matrix_instance_q31 * pDst) 
00074 { 
00075   q31_t *pIn1 = pSrcA->pData;                    /* input data matrix pointer A */ 
00076   q31_t *pIn2 = pSrcB->pData;                    /* input data matrix pointer B */ 
00077   q31_t *pInA = pSrcA->pData;                    /* input data matrix pointer A */ 
00078 //  q31_t *pSrcB = pSrcB->pData;                    /* input data matrix pointer B */  
00079   q31_t *pOut = pDst->pData;                     /* output data matrix pointer */ 
00080   q31_t *px;                                     /* Temporary output data matrix pointer */ 
00081   q63_t sum;                                     /* Accumulator */ 
00082   uint16_t numRowsA = pSrcA->numRows;            /* number of rows of input matrix A    */ 
00083   uint16_t numColsB = pSrcB->numCols;            /* number of columns of input matrix B */ 
00084   uint16_t numColsA = pSrcA->numCols;            /* number of columns of input matrix A */ 
00085   uint16_t col, i = 0u, j, row = numRowsA, colCnt;      /* loop counters */ 
00086   arm_status status;                             /* status of matrix multiplication */ 
00087  
00088  
00089 #ifdef ARM_MATH_MATRIX_CHECK 
00090   /* Check for matrix mismatch condition */ 
00091   if((pSrcA->numCols != pSrcB->numRows) || 
00092      (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols)) 
00093   { 
00094     /* Set status as ARM_MATH_SIZE_MISMATCH */ 
00095     status = ARM_MATH_SIZE_MISMATCH; 
00096   } 
00097   else 
00098 #endif 
00099   { 
00100     /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */ 
00101     /* row loop */ 
00102     do 
00103     { 
00104       /* Output pointer is set to starting address of the row being processed */ 
00105       px = pOut + i; 
00106  
00107       /* For every row wise process, the column loop counter is to be initiated */ 
00108       col = numColsB; 
00109  
00110       /* For every row wise process, the pIn2 pointer is set  
00111        ** to the starting address of the pSrcB data */ 
00112       pIn2 = pSrcB->pData; 
00113  
00114       j = 0u; 
00115  
00116       /* column loop */ 
00117       do 
00118       { 
00119         /* Set the variable sum, that acts as accumulator, to zero */ 
00120         sum = 0; 
00121  
00122         /* Initiate the pointer pIn1 to point to the starting address of pInA */ 
00123         pIn1 = pInA; 
00124  
00125         /* Apply loop unrolling and compute 4 MACs simultaneously. */ 
00126         colCnt = numColsA >> 2; 
00127  
00128  
00129         /* matrix multiplication */ 
00130         while(colCnt > 0u) 
00131         { 
00132           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ 
00133           /* Perform the multiply-accumulates */ 
00134           sum += (q63_t) * pIn1++ * *pIn2; 
00135           pIn2 += numColsB; 
00136  
00137           sum += (q63_t) * pIn1++ * *pIn2; 
00138           pIn2 += numColsB; 
00139  
00140           sum += (q63_t) * pIn1++ * *pIn2; 
00141           pIn2 += numColsB; 
00142  
00143           sum += (q63_t) * pIn1++ * *pIn2; 
00144           pIn2 += numColsB; 
00145  
00146           /* Decrement the loop counter */ 
00147           colCnt--; 
00148         } 
00149  
00150         /* If the columns of pSrcA is not a multiple of 4, compute any remaining output samples here.  
00151          ** No loop unrolling is used. */ 
00152         colCnt = numColsA % 0x4u; 
00153  
00154         while(colCnt > 0u) 
00155         { 
00156           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ 
00157           /* Perform the multiply-accumulates */ 
00158           sum += (q63_t) * pIn1++ * *pIn2; 
00159           pIn2 += numColsB; 
00160  
00161           /* Decrement the loop counter */ 
00162           colCnt--; 
00163         } 
00164  
00165         /* Convert the result from 2.30 to 1.31 format and store in destination buffer */ 
00166         *px++ = (q31_t) (sum >> 31); 
00167  
00168         /* Update the pointer pIn2 to point to the  starting address of the next column */ 
00169         j++; 
00170         pIn2 = (pSrcB->pData) + j; 
00171  
00172         /* Decrement the column loop counter */ 
00173         col--; 
00174  
00175       } while(col > 0u); 
00176  
00177       /* Update the pointer pInA to point to the  starting address of the next row */ 
00178       i = i + numColsB; 
00179       pInA = pInA + numColsA; 
00180  
00181       /* Decrement the row loop counter */ 
00182       row--; 
00183  
00184     } while(row > 0u); 
00185  
00186     /* set status as ARM_MATH_SUCCESS */ 
00187     status = ARM_MATH_SUCCESS; 
00188   } 
00189   /* Return to application */ 
00190   return (status); 
00191 } 
00192  
00193 /**  
00194  * @} end of MatrixMult group  
00195  */