tmk_core/tool/mbed/mbed-sdk/libraries/dsp/cmsis_dsp/MatrixFunctions/arm_mat_mult_f32.c

   1 /* ----------------------------------------------------------------------
   2 * Copyright (C) 2010-2013 ARM Limited. All rights reserved.
   3 *
   4 * $Date:        17. January 2013
   5 * $Revision:    V1.4.1
   6 *
   7 * Project:          CMSIS DSP Library
   8 * Title:            arm_mat_mult_f32.c
   9 *
  10 * Description:  Floating-point matrix multiplication.
  11 *
  12 * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0
  13 *
  14 * Redistribution and use in source and binary forms, with or without
  15 * modification, are permitted provided that the following conditions
  16 * are met:
  17 *   - Redistributions of source code must retain the above copyright
  18 *     notice, this list of conditions and the following disclaimer.
  19 *   - Redistributions in binary form must reproduce the above copyright
  20 *     notice, this list of conditions and the following disclaimer in
  21 *     the documentation and/or other materials provided with the
  22 *     distribution.
  23 *   - Neither the name of ARM LIMITED nor the names of its contributors
  24 *     may be used to endorse or promote products derived from this
  25 *     software without specific prior written permission.
  26 *
  27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  30 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
  31 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  32 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  33 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  34 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  35 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
  37 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  38 * POSSIBILITY OF SUCH DAMAGE.
  39 * -------------------------------------------------------------------- */
  40
  41 #include "arm_math.h"
  42
  43 /**
  44  * @ingroup groupMatrix
  45  */
  46
  47 /**
  48  * @defgroup MatrixMult Matrix Multiplication
  49  *
  50  * Multiplies two matrices.
  51  *
  52  * \image html MatrixMultiplication.gif "Multiplication of two 3 x 3 matrices"
  53
  54  * Matrix multiplication is only defined if the number of columns of the
  55  * first matrix equals the number of rows of the second matrix.
  56  * Multiplying an <code>M x N</code> matrix with an <code>N x P</code> matrix results
  57  * in an <code>M x P</code> matrix.
  58  * When matrix size checking is enabled, the functions check: (1) that the inner dimensions of
  59  * <code>pSrcA</code> and <code>pSrcB</code> are equal; and (2) that the size of the output
  60  * matrix equals the outer dimensions of <code>pSrcA</code> and <code>pSrcB</code>.
  61  */
  62
  63
  64 /**
  65  * @addtogroup MatrixMult
  66  * @{
  67  */
  68
  69 /**
  70  * @brief Floating-point matrix multiplication.
  71  * @param[in]       *pSrcA points to the first input matrix structure
  72  * @param[in]       *pSrcB points to the second input matrix structure
  73  * @param[out]      *pDst points to output matrix structure
  74  * @return              The function returns either
  75  * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
  76  */
  77
  78 arm_status arm_mat_mult_f32(
  79   const arm_matrix_instance_f32 * pSrcA,
  80   const arm_matrix_instance_f32 * pSrcB,
  81   arm_matrix_instance_f32 * pDst)
  82 {
  83   float32_t *pIn1 = pSrcA->pData;                /* input data matrix pointer A */
  84   float32_t *pIn2 = pSrcB->pData;                /* input data matrix pointer B */
  85   float32_t *pInA = pSrcA->pData;                /* input data matrix pointer A  */
  86   float32_t *pOut = pDst->pData;                 /* output data matrix pointer */
  87   float32_t *px;                                 /* Temporary output data matrix pointer */
  88   float32_t sum;                                 /* Accumulator */
  89   uint16_t numRowsA = pSrcA->numRows;            /* number of rows of input matrix A */
  90   uint16_t numColsB = pSrcB->numCols;            /* number of columns of input matrix B */
  91   uint16_t numColsA = pSrcA->numCols;            /* number of columns of input matrix A */
  92
  93 #ifndef ARM_MATH_CM0_FAMILY
  94
  95   /* Run the below code for Cortex-M4 and Cortex-M3 */
  96
  97   float32_t in1, in2, in3, in4;
  98   uint16_t col, i = 0u, j, row = numRowsA, colCnt;      /* loop counters */
  99   arm_status status;                             /* status of matrix multiplication */
 100
 101 #ifdef ARM_MATH_MATRIX_CHECK
 102
 103
 104   /* Check for matrix mismatch condition */
 105   if((pSrcA->numCols != pSrcB->numRows) ||
 106      (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols))
 107   {
 108
 109     /* Set status as ARM_MATH_SIZE_MISMATCH */
 110     status = ARM_MATH_SIZE_MISMATCH;
 111   }
 112   else
 113 #endif /*      #ifdef ARM_MATH_MATRIX_CHECK    */
 114
 115   {
 116     /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */
 117     /* row loop */
 118     do
 119     {
 120       /* Output pointer is set to starting address of the row being processed */
 121       px = pOut + i;
 122
 123       /* For every row wise process, the column loop counter is to be initiated */
 124       col = numColsB;
 125
 126       /* For every row wise process, the pIn2 pointer is set
 127        ** to the starting address of the pSrcB data */
 128       pIn2 = pSrcB->pData;
 129
 130       j = 0u;
 131
 132       /* column loop */
 133       do
 134       {
 135         /* Set the variable sum, that acts as accumulator, to zero */
 136         sum = 0.0f;
 137
 138         /* Initiate the pointer pIn1 to point to the starting address of the column being processed */
 139         pIn1 = pInA;
 140
 141         /* Apply loop unrolling and compute 4 MACs simultaneously. */
 142         colCnt = numColsA >> 2u;
 143
 144         /* matrix multiplication        */
 145         while(colCnt > 0u)
 146         {
 147           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */
 148           in3 = *pIn2;
 149           pIn2 += numColsB;
 150           in1 = pIn1[0];
 151           in2 = pIn1[1];
 152           sum += in1 * in3;
 153           in4 = *pIn2;
 154           pIn2 += numColsB;
 155           sum += in2 * in4;
 156
 157           in3 = *pIn2;
 158           pIn2 += numColsB;
 159           in1 = pIn1[2];
 160           in2 = pIn1[3];
 161           sum += in1 * in3;
 162           in4 = *pIn2;
 163           pIn2 += numColsB;
 164           sum += in2 * in4;
 165           pIn1 += 4u;
 166
 167           /* Decrement the loop count */
 168           colCnt--;
 169         }
 170
 171         /* If the columns of pSrcA is not a multiple of 4, compute any remaining MACs here.
 172          ** No loop unrolling is used. */
 173         colCnt = numColsA % 0x4u;
 174
 175         while(colCnt > 0u)
 176         {
 177           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */
 178           sum += *pIn1++ * (*pIn2);
 179           pIn2 += numColsB;
 180
 181           /* Decrement the loop counter */
 182           colCnt--;
 183         }
 184
 185         /* Store the result in the destination buffer */
 186         *px++ = sum;
 187
 188         /* Update the pointer pIn2 to point to the  starting address of the next column */
 189         j++;
 190         pIn2 = pSrcB->pData + j;
 191
 192         /* Decrement the column loop counter */
 193         col--;
 194
 195       } while(col > 0u);
 196
 197 #else
 198
 199   /* Run the below code for Cortex-M0 */
 200
 201   float32_t *pInB = pSrcB->pData;                /* input data matrix pointer B */
 202   uint16_t col, i = 0u, row = numRowsA, colCnt;  /* loop counters */
 203   arm_status status;                             /* status of matrix multiplication */
 204
 205 #ifdef ARM_MATH_MATRIX_CHECK
 206
 207   /* Check for matrix mismatch condition */
 208   if((pSrcA->numCols != pSrcB->numRows) ||
 209      (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols))
 210   {
 211
 212     /* Set status as ARM_MATH_SIZE_MISMATCH */
 213     status = ARM_MATH_SIZE_MISMATCH;
 214   }
 215   else
 216 #endif /*      #ifdef ARM_MATH_MATRIX_CHECK    */
 217
 218   {
 219     /* The following loop performs the dot-product of each row in pInA with each column in pInB */
 220     /* row loop */
 221     do
 222     {
 223       /* Output pointer is set to starting address of the row being processed */
 224       px = pOut + i;
 225
 226       /* For every row wise process, the column loop counter is to be initiated */
 227       col = numColsB;
 228
 229       /* For every row wise process, the pIn2 pointer is set
 230        ** to the starting address of the pSrcB data */
 231       pIn2 = pSrcB->pData;
 232
 233       /* column loop */
 234       do
 235       {
 236         /* Set the variable sum, that acts as accumulator, to zero */
 237         sum = 0.0f;
 238
 239         /* Initialize the pointer pIn1 to point to the starting address of the row being processed */
 240         pIn1 = pInA;
 241
 242         /* Matrix A columns number of MAC operations are to be performed */
 243         colCnt = numColsA;
 244
 245         while(colCnt > 0u)
 246         {
 247           /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */
 248           sum += *pIn1++ * (*pIn2);
 249           pIn2 += numColsB;
 250
 251           /* Decrement the loop counter */
 252           colCnt--;
 253         }
 254
 255         /* Store the result in the destination buffer */
 256         *px++ = sum;
 257
 258         /* Decrement the column loop counter */
 259         col--;
 260
 261         /* Update the pointer pIn2 to point to the  starting address of the next column */
 262         pIn2 = pInB + (numColsB - col);
 263
 264       } while(col > 0u);
 265
 266 #endif /* #ifndef ARM_MATH_CM0_FAMILY */
 267
 268       /* Update the pointer pInA to point to the  starting address of the next row */
 269       i = i + numColsB;
 270       pInA = pInA + numColsA;
 271
 272       /* Decrement the row loop counter */
 273       row--;
 274
 275     } while(row > 0u);
 276     /* Set status as ARM_MATH_SUCCESS */
 277     status = ARM_MATH_SUCCESS;
 278   }
 279
 280   /* Return to application */
 281   return (status);
 282 }
 283
 284 /**
 285  * @} end of MatrixMult group
 286  */