arm_mat_cmplx_mult_q31.c

00001 /* ----------------------------------------------------------------------      
00002 * Copyright (C) 2010-2014 ARM Limited. All rights reserved. 
00003 *      
00004 * $Date:        19. March 2015
00005 * $Revision:    V.1.4.5
00006 *      
00007 * Project:      CMSIS DSP Library 
00008 * Title:        arm_mat_cmplx_mult_q31.c      
00009 *      
00010 * Description:  Floating-point matrix multiplication.      
00011 *      
00012 * Target Processor:          Cortex-M4/Cortex-M3/Cortex-M0
00013 *
00014 * Redistribution and use in source and binary forms, with or without 
00015 * modification, are permitted provided that the following conditions
00016 * are met:
00017 *   - Redistributions of source code must retain the above copyright
00018 *     notice, this list of conditions and the following disclaimer.
00019 *   - Redistributions in binary form must reproduce the above copyright
00020 *     notice, this list of conditions and the following disclaimer in
00021 *     the documentation and/or other materials provided with the 
00022 *     distribution.
00023 *   - Neither the name of ARM LIMITED nor the names of its contributors
00024 *     may be used to endorse or promote products derived from this
00025 *     software without specific prior written permission.
00026 *
00027 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
00028 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
00029 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
00030 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 
00031 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
00032 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
00033 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
00034 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
00035 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
00036 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
00037 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
00038 * POSSIBILITY OF SUCH DAMAGE.     
00039 * -------------------------------------------------------------------- */
00040 #include "arm_math.h"
00041 
00042 /**     
00043  * @ingroup groupMatrix     
00044  */
00045 
00046 /**      
00047  * @addtogroup CmplxMatrixMult      
00048  * @{      
00049  */
00050 
00051 /**      
00052  * @brief Q31 Complex matrix multiplication      
00053  * @param[in]       *pSrcA points to the first input complex matrix structure      
00054  * @param[in]       *pSrcB points to the second input complex matrix structure      
00055  * @param[out]      *pDst points to output complex matrix structure      
00056  * @return          The function returns either      
00057  * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.      
00058  *      
00059  * @details      
00060  * <b>Scaling and Overflow Behavior:</b>      
00061  *      
00062  * \par      
00063  * The function is implemented using an internal 64-bit accumulator.      
00064  * The accumulator has a 2.62 format and maintains full precision of the intermediate      
00065  * multiplication results but provides only a single guard bit. There is no saturation      
00066  * on intermediate additions. Thus, if the accumulator overflows it wraps around and      
00067  * distorts the result. The input signals should be scaled down to avoid intermediate      
00068  * overflows. The input is thus scaled down by log2(numColsA) bits      
00069  * to avoid overflows, as a total of numColsA additions are performed internally.      
00070  * The 2.62 accumulator is right shifted by 31 bits and saturated to 1.31 format to yield the final result.      
00071  *      
00072  *      
00073  */
00074 
00075 arm_status arm_mat_cmplx_mult_q31(
00076   const arm_matrix_instance_q31 * pSrcA,
00077   const arm_matrix_instance_q31 * pSrcB,
00078   arm_matrix_instance_q31 * pDst)
00079 {
00080   q31_t *pIn1 = pSrcA->pData;                    /* input data matrix pointer A */
00081   q31_t *pIn2 = pSrcB->pData;                    /* input data matrix pointer B */
00082   q31_t *pInA = pSrcA->pData;                    /* input data matrix pointer A  */
00083   q31_t *pOut = pDst->pData;                     /* output data matrix pointer */
00084   q31_t *px;                                     /* Temporary output data matrix pointer */
00085   uint16_t numRowsA = pSrcA->numRows;            /* number of rows of input matrix A */
00086   uint16_t numColsB = pSrcB->numCols;            /* number of columns of input matrix B */
00087   uint16_t numColsA = pSrcA->numCols;            /* number of columns of input matrix A */
00088   q63_t sumReal1, sumImag1;                      /* accumulator */
00089   q31_t a0, b0, c0, d0;
00090   q31_t a1, b1, c1, d1;
00091 
00092 
00093   /* Run the below code for Cortex-M4 and Cortex-M3 */
00094 
00095   uint16_t col, i = 0u, j, row = numRowsA, colCnt;      /* loop counters */
00096   arm_status status;                             /* status of matrix multiplication */
00097 
00098 #ifdef ARM_MATH_MATRIX_CHECK
00099 
00100 
00101   /* Check for matrix mismatch condition */
00102   if((pSrcA->numCols != pSrcB->numRows) ||
00103      (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols))
00104   {
00105 
00106     /* Set status as ARM_MATH_SIZE_MISMATCH */
00107     status = ARM_MATH_SIZE_MISMATCH;
00108   }
00109   else
00110 #endif /*      #ifdef ARM_MATH_MATRIX_CHECK    */
00111 
00112   {
00113     /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */
00114     /* row loop */
00115     do
00116     {
00117       /* Output pointer is set to starting address of the row being processed */
00118       px = pOut + 2 * i;
00119 
00120       /* For every row wise process, the column loop counter is to be initiated */
00121       col = numColsB;
00122 
00123       /* For every row wise process, the pIn2 pointer is set     
00124        ** to the starting address of the pSrcB data */
00125       pIn2 = pSrcB->pData;
00126 
00127       j = 0u;
00128 
00129       /* column loop */
00130       do
00131       {
00132         /* Set the variable sum, that acts as accumulator, to zero */
00133         sumReal1 = 0.0;
00134         sumImag1 = 0.0;
00135 
00136         /* Initiate the pointer pIn1 to point to the starting address of the column being processed */
00137         pIn1 = pInA;
00138 
00139         /* Apply loop unrolling and compute 4 MACs simultaneously. */
00140         colCnt = numColsA >> 2;
00141 
00142         /* matrix multiplication        */
00143         while(colCnt > 0u)
00144         {
00145 
00146           /* Reading real part of complex matrix A */
00147           a0 = *pIn1;
00148 
00149           /* Reading real part of complex matrix B */
00150           c0 = *pIn2;
00151 
00152           /* Reading imaginary part of complex matrix A */
00153           b0 = *(pIn1 + 1u);
00154 
00155           /* Reading imaginary part of complex matrix B */
00156           d0 = *(pIn2 + 1u);
00157 
00158           /* Multiply and Accumlates */
00159           sumReal1 += (q63_t) a0 *c0;
00160           sumImag1 += (q63_t) b0 *c0;
00161 
00162           /* update pointers */
00163           pIn1 += 2u;
00164           pIn2 += 2 * numColsB;
00165 
00166           /* Multiply and Accumlates */
00167           sumReal1 -= (q63_t) b0 *d0;
00168           sumImag1 += (q63_t) a0 *d0;
00169 
00170           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */
00171 
00172           /* read real and imag values from pSrcA and pSrcB buffer */
00173           a1 = *pIn1;
00174           c1 = *pIn2;
00175           b1 = *(pIn1 + 1u);
00176           d1 = *(pIn2 + 1u);
00177 
00178           /* Multiply and Accumlates */
00179           sumReal1 += (q63_t) a1 *c1;
00180           sumImag1 += (q63_t) b1 *c1;
00181 
00182           /* update pointers */
00183           pIn1 += 2u;
00184           pIn2 += 2 * numColsB;
00185 
00186           /* Multiply and Accumlates */
00187           sumReal1 -= (q63_t) b1 *d1;
00188           sumImag1 += (q63_t) a1 *d1;
00189 
00190           a0 = *pIn1;
00191           c0 = *pIn2;
00192 
00193           b0 = *(pIn1 + 1u);
00194           d0 = *(pIn2 + 1u);
00195 
00196           /* Multiply and Accumlates */
00197           sumReal1 += (q63_t) a0 *c0;
00198           sumImag1 += (q63_t) b0 *c0;
00199 
00200           /* update pointers */
00201           pIn1 += 2u;
00202           pIn2 += 2 * numColsB;
00203 
00204           /* Multiply and Accumlates */
00205           sumReal1 -= (q63_t) b0 *d0;
00206           sumImag1 += (q63_t) a0 *d0;
00207 
00208           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */
00209 
00210           a1 = *pIn1;
00211           c1 = *pIn2;
00212 
00213           b1 = *(pIn1 + 1u);
00214           d1 = *(pIn2 + 1u);
00215 
00216           /* Multiply and Accumlates */
00217           sumReal1 += (q63_t) a1 *c1;
00218           sumImag1 += (q63_t) b1 *c1;
00219 
00220           /* update pointers */
00221           pIn1 += 2u;
00222           pIn2 += 2 * numColsB;
00223 
00224           /* Multiply and Accumlates */
00225           sumReal1 -= (q63_t) b1 *d1;
00226           sumImag1 += (q63_t) a1 *d1;
00227 
00228           /* Decrement the loop count */
00229           colCnt--;
00230         }
00231 
00232         /* If the columns of pSrcA is not a multiple of 4, compute any remaining MACs here.     
00233          ** No loop unrolling is used. */
00234         colCnt = numColsA % 0x4u;
00235 
00236         while(colCnt > 0u)
00237         {
00238           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */
00239           a1 = *pIn1;
00240           c1 = *pIn2;
00241 
00242           b1 = *(pIn1 + 1u);
00243           d1 = *(pIn2 + 1u);
00244 
00245           /* Multiply and Accumlates */
00246           sumReal1 += (q63_t) a1 *c1;
00247           sumImag1 += (q63_t) b1 *c1;
00248 
00249           /* update pointers */
00250           pIn1 += 2u;
00251           pIn2 += 2 * numColsB;
00252 
00253           /* Multiply and Accumlates */
00254           sumReal1 -= (q63_t) b1 *d1;
00255           sumImag1 += (q63_t) a1 *d1;
00256 
00257           /* Decrement the loop counter */
00258           colCnt--;
00259         }
00260 
00261         /* Store the result in the destination buffer */
00262         *px++ = (q31_t) clip_q63_to_q31(sumReal1 >> 31);
00263         *px++ = (q31_t) clip_q63_to_q31(sumImag1 >> 31);
00264         
00265         /* Update the pointer pIn2 to point to the  starting address of the next column */
00266         j++;
00267         pIn2 = pSrcB->pData + 2u * j;
00268 
00269         /* Decrement the column loop counter */
00270         col--;
00271 
00272       } while(col > 0u);
00273 
00274       /* Update the pointer pInA to point to the  starting address of the next row */
00275       i = i + numColsB;
00276       pInA = pInA + 2 * numColsA;
00277 
00278       /* Decrement the row loop counter */
00279       row--;
00280 
00281     } while(row > 0u);
00282 
00283     /* Set status as ARM_MATH_SUCCESS */
00284     status = ARM_MATH_SUCCESS;
00285   }
00286 
00287   /* Return to application */
00288   return (status);
00289 }
00290 
00291 /**     
00292  * @} end of MatrixMult group     
00293  */