arm_mat_mult_f32.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_f32.c  
00009 *  
00010 * Description:  Floating-point 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  * @defgroup MatrixMult Matrix Multiplication  
00041  *  
00042  * Multiplies two matrices.  
00043  *  
00044  * \image html MatrixMultiplication.gif "Multiplication of two 3 x 3 matrices"  
00045   
00046  * Matrix multiplication is only defined if the number of columns of the  
00047  * first matrix equals the number of rows of the second matrix.  
00048  * Multiplying an <code>M x N</code> matrix with an <code>N x P</code> matrix results  
00049  * in an <code>M x P</code> matrix.  
00050  * When matrix size checking is enabled, the functions check: (1) that the inner dimensions of  
00051  * <code>pSrcA</code> and <code>pSrcB</code> are equal; and (2) that the size of the output  
00052  * matrix equals the outer dimensions of <code>pSrcA</code> and <code>pSrcB</code>.  
00053  */ 
00054  
00055  
00056 /**  
00057  * @addtogroup MatrixMult  
00058  * @{  
00059  */ 
00060  
00061 /**  
00062  * @brief Floating-point matrix multiplication.  
00063  * @param[in]       *pSrcA points to the first input matrix structure  
00064  * @param[in]       *pSrcB points to the second input matrix structure  
00065  * @param[out]      *pDst points to output matrix structure  
00066  * @return          The function returns either  
00067  * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.  
00068  */ 
00069  
00070 arm_status arm_mat_mult_f32( 
00071   const arm_matrix_instance_f32 * pSrcA, 
00072   const arm_matrix_instance_f32 * pSrcB, 
00073   arm_matrix_instance_f32 * pDst) 
00074 { 
00075   float32_t *pIn1 = pSrcA->pData;                /* input data matrix pointer A */ 
00076   float32_t *pIn2 = pSrcB->pData;                /* input data matrix pointer B */ 
00077   float32_t *pInA = pSrcA->pData;                /* input data matrix pointer A  */ 
00078 //  float32_t *pSrcB = pSrcB->pData;                /* input data matrix pointer B */  
00079   float32_t *pOut = pDst->pData;                 /* output data matrix pointer */ 
00080   float32_t *px;                                 /* Temporary output data matrix pointer */ 
00081   float32_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 #ifdef ARM_MATH_MATRIX_CHECK 
00089   /* Check for matrix mismatch condition */ 
00090   if((pSrcA->numCols != pSrcB->numRows) || 
00091      (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols)) 
00092   { 
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 = 0.0f; 
00121  
00122         /* Initiate the pointer pIn1 to point to the starting address of the column being processed */ 
00123         pIn1 = pInA; 
00124  
00125         /* Apply loop unrolling and compute 4 MACs simultaneously. */ 
00126         colCnt = numColsA >> 2; 
00127  
00128         /* matrix multiplication        */ 
00129         while(colCnt > 0u) 
00130         { 
00131           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ 
00132           sum += *pIn1++ * (*pIn2); 
00133           pIn2 += numColsB; 
00134           sum += *pIn1++ * (*pIn2); 
00135           pIn2 += numColsB; 
00136           sum += *pIn1++ * (*pIn2); 
00137           pIn2 += numColsB; 
00138           sum += *pIn1++ * (*pIn2); 
00139           pIn2 += numColsB; 
00140  
00141           /* Decrement the loop count */ 
00142           colCnt--; 
00143         } 
00144  
00145         /* If the columns of pSrcA is not a multiple of 4, compute any remaining MACs here.  
00146          ** No loop unrolling is used. */ 
00147         colCnt = numColsA % 0x4u; 
00148  
00149         while(colCnt > 0u) 
00150         { 
00151           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ 
00152           sum += *pIn1++ * (*pIn2); 
00153           pIn2 += numColsB; 
00154  
00155           /* Decrement the loop counter */ 
00156           colCnt--; 
00157         } 
00158  
00159         /* Store the result in the destination buffer */ 
00160         *px++ = sum; 
00161  
00162         /* Update the pointer pIn2 to point to the  starting address of the next column */ 
00163         j++; 
00164         pIn2 = pSrcB->pData + j; 
00165  
00166         /* Decrement the column loop counter */ 
00167         col--; 
00168  
00169       } while(col > 0u); 
00170  
00171       /* Update the pointer pInA to point to the  starting address of the next row */ 
00172       i = i + numColsB; 
00173       pInA = pInA + numColsA; 
00174  
00175       /* Decrement the row loop counter */ 
00176       row--; 
00177  
00178     } while(row > 0u); 
00179  
00180     /* Set status as ARM_MATH_SUCCESS */ 
00181     status = ARM_MATH_SUCCESS; 
00182   } 
00183  
00184   /* Return to application */ 
00185   return (status); 
00186 } 
00187  
00188 /**  
00189  * @} end of MatrixMult group  
00190  */