arm_mat_cmplx_mult_q15.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_cmplx_mat_mult_q15.c      
00009 *      
00010 * Description:   Q15 complex 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 /**      
00053  * @brief Q15 Complex matrix multiplication      
00054  * @param[in]       *pSrcA points to the first input complex matrix structure      
00055  * @param[in]       *pSrcB points to the second input complex matrix structure      
00056  * @param[out]      *pDst points to output complex matrix structure      
00057  * @param[in]       *pScratch points to the array for storing intermediate results     
00058  * @return          The function returns either      
00059  * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.         
00060  *  
00061  * \par Conditions for optimum performance  
00062  *  Input, output and state buffers should be aligned by 32-bit  
00063  *  
00064  * \par Restrictions  
00065  *  If the silicon does not support unaligned memory access enable the macro UNALIGNED_SUPPORT_DISABLE  
00066  *  In this case input, output, scratch buffers should be aligned by 32-bit  
00067  *  
00068  * @details      
00069  * <b>Scaling and Overflow Behavior:</b>      
00070  *      
00071  * \par      
00072  * The function is implemented using a 64-bit internal accumulator. The inputs to the      
00073  * multiplications are in 1.15 format and multiplications yield a 2.30 result.      
00074  * The 2.30 intermediate      
00075  * results are accumulated in a 64-bit accumulator in 34.30 format. This approach      
00076  * provides 33 guard bits and there is no risk of overflow. The 34.30 result is then      
00077  * truncated to 34.15 format by discarding the low 15 bits and then saturated to      
00078  * 1.15 format.      
00079  *      
00080  * \par      
00081  * Refer to <code>arm_mat_mult_fast_q15()</code> for a faster but less precise version of this function.      
00082  *      
00083  */
00084 
00085 
00086 
00087 
00088 arm_status arm_mat_cmplx_mult_q15(
00089   const arm_matrix_instance_q15 * pSrcA,
00090   const arm_matrix_instance_q15 * pSrcB,
00091   arm_matrix_instance_q15 * pDst,
00092   q15_t * pScratch)
00093 {
00094   /* accumulator */
00095   q15_t *pSrcBT = pScratch;                      /* input data matrix pointer for transpose */
00096   q15_t *pInA = pSrcA->pData;                    /* input data matrix pointer A of Q15 type */
00097   q15_t *pInB = pSrcB->pData;                    /* input data matrix pointer B of Q15 type */
00098   q15_t *px;                                     /* Temporary output data matrix pointer */
00099   uint16_t numRowsA = pSrcA->numRows;            /* number of rows of input matrix A    */
00100   uint16_t numColsB = pSrcB->numCols;            /* number of columns of input matrix B */
00101   uint16_t numColsA = pSrcA->numCols;            /* number of columns of input matrix A */
00102   uint16_t numRowsB = pSrcB->numRows;            /* number of rows of input matrix A    */
00103   uint16_t col, i = 0u, row = numRowsB, colCnt;  /* loop counters */
00104   arm_status status;                             /* status of matrix multiplication */
00105   q63_t sumReal, sumImag;
00106 
00107 #ifdef UNALIGNED_SUPPORT_DISABLE
00108   q15_t in;                                      /* Temporary variable to hold the input value */
00109   q15_t a, b, c, d;
00110 #else
00111   q31_t in;                                      /* Temporary variable to hold the input value */
00112   q31_t prod1, prod2;
00113   q31_t pSourceA, pSourceB;
00114 #endif
00115 
00116 #ifdef ARM_MATH_MATRIX_CHECK
00117   /* Check for matrix mismatch condition */
00118   if((pSrcA->numCols != pSrcB->numRows) ||
00119      (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols))
00120   {
00121     /* Set status as ARM_MATH_SIZE_MISMATCH */
00122     status = ARM_MATH_SIZE_MISMATCH;
00123   }
00124   else
00125 #endif
00126   {
00127     /* Matrix transpose */
00128     do
00129     {
00130       /* Apply loop unrolling and exchange the columns with row elements */
00131       col = numColsB >> 2;
00132 
00133       /* The pointer px is set to starting address of the column being processed */
00134       px = pSrcBT + i;
00135 
00136       /* First part of the processing with loop unrolling.  Compute 4 outputs at a time.      
00137        ** a second loop below computes the remaining 1 to 3 samples. */
00138       while(col > 0u)
00139       {
00140 #ifdef UNALIGNED_SUPPORT_DISABLE
00141         /* Read two elements from the row */
00142         in = *pInB++;
00143         *px = in;
00144         in = *pInB++;
00145         px[1] = in;
00146 
00147         /* Update the pointer px to point to the next row of the transposed matrix */
00148         px += numRowsB * 2;
00149 
00150         /* Read two elements from the row */
00151         in = *pInB++;
00152         *px = in;
00153         in = *pInB++;
00154         px[1] = in;
00155 
00156         /* Update the pointer px to point to the next row of the transposed matrix */
00157         px += numRowsB * 2;
00158 
00159         /* Read two elements from the row */
00160         in = *pInB++;
00161         *px = in;
00162         in = *pInB++;
00163         px[1] = in;
00164 
00165         /* Update the pointer px to point to the next row of the transposed matrix */
00166         px += numRowsB * 2;
00167 
00168         /* Read two elements from the row */
00169         in = *pInB++;
00170         *px = in;
00171         in = *pInB++;
00172         px[1] = in;
00173 
00174         /* Update the pointer px to point to the next row of the transposed matrix */
00175         px += numRowsB * 2;
00176 
00177         /* Decrement the column loop counter */
00178         col--;
00179       }
00180 
00181       /* If the columns of pSrcB is not a multiple of 4, compute any remaining output samples here.      
00182        ** No loop unrolling is used. */
00183       col = numColsB % 0x4u;
00184 
00185       while(col > 0u)
00186       {
00187         /* Read two elements from the row */
00188         in = *pInB++;
00189         *px = in;
00190         in = *pInB++;
00191         px[1] = in;
00192 #else
00193 
00194         /* Read two elements from the row */
00195         in = *__SIMD32(pInB)++;
00196 
00197         *__SIMD32(px) = in;
00198 
00199         /* Update the pointer px to point to the next row of the transposed matrix */
00200         px += numRowsB * 2;
00201 
00202 
00203         /* Read two elements from the row */
00204         in = *__SIMD32(pInB)++;
00205 
00206         *__SIMD32(px) = in;
00207 
00208         /* Update the pointer px to point to the next row of the transposed matrix */
00209         px += numRowsB * 2;
00210 
00211         /* Read two elements from the row */
00212         in = *__SIMD32(pInB)++;
00213 
00214         *__SIMD32(px) = in;
00215 
00216         /* Update the pointer px to point to the next row of the transposed matrix */
00217         px += numRowsB * 2;
00218 
00219         /* Read two elements from the row */
00220         in = *__SIMD32(pInB)++;
00221 
00222         *__SIMD32(px) = in;
00223 
00224         /* Update the pointer px to point to the next row of the transposed matrix */
00225         px += numRowsB * 2;
00226 
00227         /* Decrement the column loop counter */
00228         col--;
00229       }
00230 
00231       /* If the columns of pSrcB is not a multiple of 4, compute any remaining output samples here.      
00232        ** No loop unrolling is used. */
00233       col = numColsB % 0x4u;
00234 
00235       while(col > 0u)
00236       {
00237         /* Read two elements from the row */
00238         in = *__SIMD32(pInB)++;
00239 
00240         *__SIMD32(px) = in;
00241 #endif
00242 
00243         /* Update the pointer px to point to the next row of the transposed matrix */
00244         px += numRowsB * 2;
00245 
00246         /* Decrement the column loop counter */
00247         col--;
00248       }
00249 
00250       i = i + 2u;
00251 
00252       /* Decrement the row loop counter */
00253       row--;
00254 
00255     } while(row > 0u);
00256 
00257     /* Reset the variables for the usage in the following multiplication process */
00258     row = numRowsA;
00259     i = 0u;
00260     px = pDst->pData;
00261 
00262     /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */
00263     /* row loop */
00264     do
00265     {
00266       /* For every row wise process, the column loop counter is to be initiated */
00267       col = numColsB;
00268 
00269       /* For every row wise process, the pIn2 pointer is set      
00270        ** to the starting address of the transposed pSrcB data */
00271       pInB = pSrcBT;
00272 
00273       /* column loop */
00274       do
00275       {
00276         /* Set the variable sum, that acts as accumulator, to zero */
00277         sumReal = 0;
00278         sumImag = 0;
00279 
00280         /* Apply loop unrolling and compute 2 MACs simultaneously. */
00281         colCnt = numColsA >> 1;
00282 
00283         /* Initiate the pointer pIn1 to point to the starting address of the column being processed */
00284         pInA = pSrcA->pData + i * 2;
00285 
00286 
00287         /* matrix multiplication */
00288         while(colCnt > 0u)
00289         {
00290           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */
00291 
00292 #ifdef UNALIGNED_SUPPORT_DISABLE
00293 
00294           /* read real and imag values from pSrcA buffer */
00295           a = *pInA;
00296           b = *(pInA + 1u);
00297           /* read real and imag values from pSrcB buffer */
00298           c = *pInB;
00299           d = *(pInB + 1u);
00300 
00301           /* Multiply and Accumlates */
00302           sumReal += (q31_t) a *c;
00303           sumImag += (q31_t) a *d;
00304           sumReal -= (q31_t) b *d;
00305           sumImag += (q31_t) b *c;
00306 
00307           /* read next real and imag values from pSrcA buffer */
00308           a = *(pInA + 2u);
00309           b = *(pInA + 3u);
00310           /* read next real and imag values from pSrcB buffer */
00311           c = *(pInB + 2u);
00312           d = *(pInB + 3u);
00313 
00314           /* update pointer */
00315           pInA += 4u;
00316 
00317           /* Multiply and Accumlates */
00318           sumReal += (q31_t) a *c;
00319           sumImag += (q31_t) a *d;
00320           sumReal -= (q31_t) b *d;
00321           sumImag += (q31_t) b *c;
00322           /* update pointer */
00323           pInB += 4u;
00324 #else
00325           /* read real and imag values from pSrcA and pSrcB buffer */
00326           pSourceA = *__SIMD32(pInA)++;
00327           pSourceB = *__SIMD32(pInB)++;
00328 
00329           /* Multiply and Accumlates */
00330 #ifdef ARM_MATH_BIG_ENDIAN
00331           prod1 = -__SMUSD(pSourceA, pSourceB);
00332 #else
00333           prod1 = __SMUSD(pSourceA, pSourceB);
00334 #endif
00335           prod2 = __SMUADX(pSourceA, pSourceB);
00336           sumReal += (q63_t) prod1;
00337           sumImag += (q63_t) prod2;
00338 
00339           /* read real and imag values from pSrcA and pSrcB buffer */
00340           pSourceA = *__SIMD32(pInA)++;
00341           pSourceB = *__SIMD32(pInB)++;
00342 
00343           /* Multiply and Accumlates */
00344 #ifdef ARM_MATH_BIG_ENDIAN
00345           prod1 = -__SMUSD(pSourceA, pSourceB);
00346 #else
00347           prod1 = __SMUSD(pSourceA, pSourceB);
00348 #endif
00349           prod2 = __SMUADX(pSourceA, pSourceB);
00350           sumReal += (q63_t) prod1;
00351           sumImag += (q63_t) prod2;
00352 
00353 #endif /*      #ifdef UNALIGNED_SUPPORT_DISABLE */
00354 
00355           /* Decrement the loop counter */
00356           colCnt--;
00357         }
00358 
00359         /* process odd column samples */
00360         if((numColsA & 0x1u) > 0u)
00361         {
00362           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */
00363 
00364 #ifdef UNALIGNED_SUPPORT_DISABLE
00365 
00366           /* read real and imag values from pSrcA and pSrcB buffer */
00367           a = *pInA++;
00368           b = *pInA++;
00369           c = *pInB++;
00370           d = *pInB++;
00371 
00372           /* Multiply and Accumlates */
00373           sumReal += (q31_t) a *c;
00374           sumImag += (q31_t) a *d;
00375           sumReal -= (q31_t) b *d;
00376           sumImag += (q31_t) b *c;
00377 
00378 #else
00379           /* read real and imag values from pSrcA and pSrcB buffer */
00380           pSourceA = *__SIMD32(pInA)++;
00381           pSourceB = *__SIMD32(pInB)++;
00382 
00383           /* Multiply and Accumlates */
00384 #ifdef ARM_MATH_BIG_ENDIAN
00385           prod1 = -__SMUSD(pSourceA, pSourceB);
00386 #else
00387           prod1 = __SMUSD(pSourceA, pSourceB);
00388 #endif
00389           prod2 = __SMUADX(pSourceA, pSourceB);
00390           sumReal += (q63_t) prod1;
00391           sumImag += (q63_t) prod2;
00392 
00393 #endif /*      #ifdef UNALIGNED_SUPPORT_DISABLE */
00394 
00395         }
00396 
00397         /* Saturate and store the result in the destination buffer */
00398 
00399         *px++ = (q15_t) (__SSAT(sumReal >> 15, 16));
00400         *px++ = (q15_t) (__SSAT(sumImag >> 15, 16));
00401 
00402         /* Decrement the column loop counter */
00403         col--;
00404 
00405       } while(col > 0u);
00406 
00407       i = i + numColsA;
00408 
00409       /* Decrement the row loop counter */
00410       row--;
00411 
00412     } while(row > 0u);
00413 
00414     /* set status as ARM_MATH_SUCCESS */
00415     status = ARM_MATH_SUCCESS;
00416   }
00417 
00418   /* Return to application */
00419   return (status);
00420 }
00421 
00422 /**      
00423  * @} end of MatrixMult group      
00424  */