| /* ---------------------------------------------------------------------- |
| * Copyright (C) 2010 ARM Limited. All rights reserved. |
| * |
| * $Date: 15. July 2011 |
| * $Revision: V1.0.10 |
| * |
| * Project: CMSIS DSP Library |
| * Title: arm_mat_mult_f32.c |
| * |
| * Description: Floating-point matrix multiplication. |
| * |
| * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0 |
| * |
| * Version 1.0.10 2011/7/15 |
| * Big Endian support added and Merged M0 and M3/M4 Source code. |
| * |
| * Version 1.0.3 2010/11/29 |
| * Re-organized the CMSIS folders and updated documentation. |
| * |
| * Version 1.0.2 2010/11/11 |
| * Documentation updated. |
| * |
| * Version 1.0.1 2010/10/05 |
| * Production release and review comments incorporated. |
| * |
| * Version 1.0.0 2010/09/20 |
| * Production release and review comments incorporated. |
| * |
| * Version 0.0.5 2010/04/26 |
| * incorporated review comments and updated with latest CMSIS layer |
| * |
| * Version 0.0.3 2010/03/10 |
| * Initial version |
| * -------------------------------------------------------------------- */ |
| |
| #include "arm_math.h" |
| |
| /** |
| * @ingroup groupMatrix |
| */ |
| |
| /** |
| * @defgroup MatrixMult Matrix Multiplication |
| * |
| * Multiplies two matrices. |
| * |
| * \image html MatrixMultiplication.gif "Multiplication of two 3 x 3 matrices" |
| |
| * Matrix multiplication is only defined if the number of columns of the |
| * first matrix equals the number of rows of the second matrix. |
| * Multiplying an <code>M x N</code> matrix with an <code>N x P</code> matrix results |
| * in an <code>M x P</code> matrix. |
| * When matrix size checking is enabled, the functions check: (1) that the inner dimensions of |
| * <code>pSrcA</code> and <code>pSrcB</code> are equal; and (2) that the size of the output |
| * matrix equals the outer dimensions of <code>pSrcA</code> and <code>pSrcB</code>. |
| */ |
| |
| |
| /** |
| * @addtogroup MatrixMult |
| * @{ |
| */ |
| |
| /** |
| * @brief Floating-point matrix multiplication. |
| * @param[in] *pSrcA points to the first input matrix structure |
| * @param[in] *pSrcB points to the second input matrix structure |
| * @param[out] *pDst points to output matrix structure |
| * @return The function returns either |
| * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking. |
| */ |
| |
| arm_status arm_mat_mult_f32( |
| const arm_matrix_instance_f32 * pSrcA, |
| const arm_matrix_instance_f32 * pSrcB, |
| arm_matrix_instance_f32 * pDst) |
| { |
| float32_t *pIn1 = pSrcA->pData; /* input data matrix pointer A */ |
| float32_t *pIn2 = pSrcB->pData; /* input data matrix pointer B */ |
| float32_t *pInA = pSrcA->pData; /* input data matrix pointer A */ |
| float32_t *pOut = pDst->pData; /* output data matrix pointer */ |
| float32_t *px; /* Temporary output data matrix pointer */ |
| float32_t sum; /* Accumulator */ |
| uint16_t numRowsA = pSrcA->numRows; /* number of rows of input matrix A */ |
| uint16_t numColsB = pSrcB->numCols; /* number of columns of input matrix B */ |
| uint16_t numColsA = pSrcA->numCols; /* number of columns of input matrix A */ |
| |
| #ifndef ARM_MATH_CM0 |
| |
| /* Run the below code for Cortex-M4 and Cortex-M3 */ |
| |
| uint16_t col, i = 0u, j, row = numRowsA, colCnt; /* loop counters */ |
| arm_status status; /* status of matrix multiplication */ |
| |
| #ifdef ARM_MATH_MATRIX_CHECK |
| |
| |
| /* Check for matrix mismatch condition */ |
| if((pSrcA->numCols != pSrcB->numRows) || |
| (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols)) |
| { |
| |
| /* Set status as ARM_MATH_SIZE_MISMATCH */ |
| status = ARM_MATH_SIZE_MISMATCH; |
| } |
| else |
| #endif /* #ifdef ARM_MATH_MATRIX_CHECK */ |
| |
| { |
| /* The following loop performs the dot-product of each row in pSrcA with each column in pSrcB */ |
| /* row loop */ |
| do |
| { |
| /* Output pointer is set to starting address of the row being processed */ |
| px = pOut + i; |
| |
| /* For every row wise process, the column loop counter is to be initiated */ |
| col = numColsB; |
| |
| /* For every row wise process, the pIn2 pointer is set |
| ** to the starting address of the pSrcB data */ |
| pIn2 = pSrcB->pData; |
| |
| j = 0u; |
| |
| /* column loop */ |
| do |
| { |
| /* Set the variable sum, that acts as accumulator, to zero */ |
| sum = 0.0f; |
| |
| /* Initiate the pointer pIn1 to point to the starting address of the column being processed */ |
| pIn1 = pInA; |
| |
| /* Apply loop unrolling and compute 4 MACs simultaneously. */ |
| colCnt = numColsA >> 2; |
| |
| /* matrix multiplication */ |
| while(colCnt > 0u) |
| { |
| /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
| sum += *pIn1++ * (*pIn2); |
| pIn2 += numColsB; |
| sum += *pIn1++ * (*pIn2); |
| pIn2 += numColsB; |
| sum += *pIn1++ * (*pIn2); |
| pIn2 += numColsB; |
| sum += *pIn1++ * (*pIn2); |
| pIn2 += numColsB; |
| |
| /* Decrement the loop count */ |
| colCnt--; |
| } |
| |
| /* If the columns of pSrcA is not a multiple of 4, compute any remaining MACs here. |
| ** No loop unrolling is used. */ |
| colCnt = numColsA % 0x4u; |
| |
| while(colCnt > 0u) |
| { |
| /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
| sum += *pIn1++ * (*pIn2); |
| pIn2 += numColsB; |
| |
| /* Decrement the loop counter */ |
| colCnt--; |
| } |
| |
| /* Store the result in the destination buffer */ |
| *px++ = sum; |
| |
| /* Update the pointer pIn2 to point to the starting address of the next column */ |
| j++; |
| pIn2 = pSrcB->pData + j; |
| |
| /* Decrement the column loop counter */ |
| col--; |
| |
| } while(col > 0u); |
| |
| #else |
| |
| /* Run the below code for Cortex-M0 */ |
| |
| float32_t *pInB = pSrcB->pData; /* input data matrix pointer B */ |
| uint16_t col, i = 0u, row = numRowsA, colCnt; /* loop counters */ |
| arm_status status; /* status of matrix multiplication */ |
| |
| #ifdef ARM_MATH_MATRIX_CHECK |
| |
| /* Check for matrix mismatch condition */ |
| if((pSrcA->numCols != pSrcB->numRows) || |
| (pSrcA->numRows != pDst->numRows) || (pSrcB->numCols != pDst->numCols)) |
| { |
| |
| /* Set status as ARM_MATH_SIZE_MISMATCH */ |
| status = ARM_MATH_SIZE_MISMATCH; |
| } |
| else |
| #endif /* #ifdef ARM_MATH_MATRIX_CHECK */ |
| |
| { |
| /* The following loop performs the dot-product of each row in pInA with each column in pInB */ |
| /* row loop */ |
| do |
| { |
| /* Output pointer is set to starting address of the row being processed */ |
| px = pOut + i; |
| |
| /* For every row wise process, the column loop counter is to be initiated */ |
| col = numColsB; |
| |
| /* For every row wise process, the pIn2 pointer is set |
| ** to the starting address of the pSrcB data */ |
| pIn2 = pSrcB->pData; |
| |
| /* column loop */ |
| do |
| { |
| /* Set the variable sum, that acts as accumulator, to zero */ |
| sum = 0.0f; |
| |
| /* Initialize the pointer pIn1 to point to the starting address of the row being processed */ |
| pIn1 = pInA; |
| |
| /* Matrix A columns number of MAC operations are to be performed */ |
| colCnt = numColsA; |
| |
| while(colCnt > 0u) |
| { |
| /* c(m,n) = a(1,1)*b(1,1) + a(1,2) * b(2,1) + .... + a(m,p)*b(p,n) */ |
| sum += *pIn1++ * (*pIn2); |
| pIn2 += numColsB; |
| |
| /* Decrement the loop counter */ |
| colCnt--; |
| } |
| |
| /* Store the result in the destination buffer */ |
| *px++ = sum; |
| |
| /* Decrement the column loop counter */ |
| col--; |
| |
| /* Update the pointer pIn2 to point to the starting address of the next column */ |
| pIn2 = pInB + (numColsB - col); |
| |
| } while(col > 0u); |
| |
| #endif /* #ifndef ARM_MATH_CM0 */ |
| |
| /* Update the pointer pInA to point to the starting address of the next row */ |
| i = i + numColsB; |
| pInA = pInA + numColsA; |
| |
| /* Decrement the row loop counter */ |
| row--; |
| |
| } while(row > 0u); |
| /* Set status as ARM_MATH_SUCCESS */ |
| status = ARM_MATH_SUCCESS; |
| } |
| |
| /* Return to application */ |
| return (status); |
| } |
| |
| /** |
| * @} end of MatrixMult group |
| */ |