diff options
Diffstat (limited to 'DSP/Source/MatrixFunctions/arm_mat_mult_f32.c')
| -rw-r--r-- | DSP/Source/MatrixFunctions/arm_mat_mult_f32.c | 372 |
1 files changed, 316 insertions, 56 deletions
diff --git a/DSP/Source/MatrixFunctions/arm_mat_mult_f32.c b/DSP/Source/MatrixFunctions/arm_mat_mult_f32.c index a038f2f..ffddf99 100644 --- a/DSP/Source/MatrixFunctions/arm_mat_mult_f32.c +++ b/DSP/Source/MatrixFunctions/arm_mat_mult_f32.c @@ -3,13 +3,13 @@ * Title: arm_mat_mult_f32.c * Description: Floating-point matrix multiplication * - * $Date: 27. January 2017 - * $Revision: V.1.5.1 + * $Date: 18. March 2019 + * $Revision: V1.6.0 * * Target Processor: Cortex-M cores * -------------------------------------------------------------------- */ /* - * Copyright (C) 2010-2017 ARM Limited or its affiliates. All rights reserved. + * Copyright (C) 2010-2019 ARM Limited or its affiliates. All rights reserved. * * SPDX-License-Identifier: Apache-2.0 * @@ -62,6 +62,9 @@ * @return The function returns either * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking. */ +#if defined(ARM_MATH_NEON) + +#define GROUPOFROWS 8 arm_status arm_mat_mult_f32( const arm_matrix_instance_f32 * pSrcA, @@ -78,32 +81,225 @@ arm_status arm_mat_mult_f32( uint16_t numColsB = pSrcB->numCols; /* number of columns of input matrix B */ uint16_t numColsA = pSrcA->numCols; /* number of columns of input matrix A */ -#if defined (ARM_MATH_DSP) - - /* Run the below code for Cortex-M4 and Cortex-M3 */ float32_t in1, in2, in3, in4; - uint16_t col, i = 0U, j, row = numRowsA, colCnt; /* loop counters */ + uint16_t col, i = 0U, j, row = numRowsA, rowCnt, colCnt; /* loop counters */ arm_status status; /* status of matrix multiplication */ -#ifdef ARM_MATH_MATRIX_CHECK + float32x4_t a0V, a1V, a2V, a3V, a4V, a5V, a6V, a7V; + float32x4_t acc0,acc1,acc2,acc3,acc4,acc5,acc6,acc7,temp; + float32x2_t accum = vdup_n_f32(0); + float32_t *pIn1B = pSrcA->pData; + float32_t *pIn1C = pSrcA->pData; + float32_t *pIn1D = pSrcA->pData; + float32_t *pIn1E = pSrcA->pData; + float32_t *pIn1F = pSrcA->pData; + float32_t *pIn1G = pSrcA->pData; + float32_t *pIn1H = pSrcA->pData; + float32_t *pxB,*pxC, *pxD, *pxE, *pxF, *pxG, *pxH; /* Temporary output data matrix pointer */ + float32_t sum0,sum1, sum2,sum3, sum4, sum5 , sum6, sum7; + +#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 + /* Row loop */ + rowCnt = row >> 3; + + while(rowCnt > 0) + { + /* Output pointer is set to starting address of the row being processed */ + px = pOut + GROUPOFROWS*i; + pxB = px + numColsB; + pxC = px + 2*numColsB; + pxD = px + 3*numColsB; + pxE = px + 4*numColsB; + pxF = px + 5*numColsB; + pxG = px + 6*numColsB; + pxH = px + 7*numColsB; + + /* 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 */ + sum0 = 0.0f; + sum1 = 0.0f; + sum2 = 0.0f; + sum3 = 0.0f; + sum4 = 0.0f; + sum5 = 0.0f; + sum6 = 0.0f; + sum7 = 0.0f; + + /* Initiate the pointer pIn1 to point to the starting address of the column being processed */ + pIn1 = pInA; + pIn1B = pIn1 + numColsA; + pIn1C = pIn1 + 2*numColsA; + pIn1D = pIn1 + 3*numColsA; + pIn1E = pIn1 + 4*numColsA; + pIn1F = pIn1 + 5*numColsA; + pIn1G = pIn1 + 6*numColsA; + pIn1H = pIn1 + 7*numColsA; + + acc0 = vdupq_n_f32(0.0); + acc1 = vdupq_n_f32(0.0); + acc2 = vdupq_n_f32(0.0); + acc3 = vdupq_n_f32(0.0); + acc4 = vdupq_n_f32(0.0); + acc5 = vdupq_n_f32(0.0); + acc6 = vdupq_n_f32(0.0); + acc7 = vdupq_n_f32(0.0); + + /* Compute 4 MACs simultaneously. */ + colCnt = numColsA >> 2U; + + /* 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) */ + a0V = vld1q_f32(pIn1); + a1V = vld1q_f32(pIn1B); + a2V = vld1q_f32(pIn1C); + a3V = vld1q_f32(pIn1D); + a4V = vld1q_f32(pIn1E); + a5V = vld1q_f32(pIn1F); + a6V = vld1q_f32(pIn1G); + a7V = vld1q_f32(pIn1H); + + pIn1 += 4; + pIn1B += 4; + pIn1C += 4; + pIn1D += 4; + pIn1E += 4; + pIn1F += 4; + pIn1G += 4; + pIn1H += 4; + + temp[0] = *pIn2; + pIn2 += numColsB; + temp[1] = *pIn2; + pIn2 += numColsB; + temp[2] = *pIn2; + pIn2 += numColsB; + temp[3] = *pIn2; + pIn2 += numColsB; + + acc0 = vmlaq_f32(acc0,a0V,temp); + acc1 = vmlaq_f32(acc1,a1V,temp); + acc2 = vmlaq_f32(acc2,a2V,temp); + acc3 = vmlaq_f32(acc3,a3V,temp); + acc4 = vmlaq_f32(acc4,a4V,temp); + acc5 = vmlaq_f32(acc5,a5V,temp); + acc6 = vmlaq_f32(acc6,a6V,temp); + acc7 = vmlaq_f32(acc7,a7V,temp); + + /* Decrement the loop count */ + colCnt--; + } + + accum = vpadd_f32(vget_low_f32(acc0), vget_high_f32(acc0)); + sum0 += accum[0] + accum[1]; + + accum = vpadd_f32(vget_low_f32(acc1), vget_high_f32(acc1)); + sum1 += accum[0] + accum[1]; + + accum = vpadd_f32(vget_low_f32(acc2), vget_high_f32(acc2)); + sum2 += accum[0] + accum[1]; + + accum = vpadd_f32(vget_low_f32(acc3), vget_high_f32(acc3)); + sum3 += accum[0] + accum[1]; + + accum = vpadd_f32(vget_low_f32(acc4), vget_high_f32(acc4)); + sum4 += accum[0] + accum[1]; + + accum = vpadd_f32(vget_low_f32(acc5), vget_high_f32(acc5)); + sum5 += accum[0] + accum[1]; + + accum = vpadd_f32(vget_low_f32(acc6), vget_high_f32(acc6)); + sum6 += accum[0] + accum[1]; + + accum = vpadd_f32(vget_low_f32(acc7), vget_high_f32(acc7)); + sum7 += accum[0] + accum[1]; + + /* If the columns of pSrcA is not a multiple of 4, compute any remaining MACs here. + ** No loop unrolling is used. */ + colCnt = numColsA & 3; + + while (colCnt > 0U) + { + /* c(m,n) = a(1,1)*b(1,1) + a(1,2)*b(2,1) + ... + a(m,p)*b(p,n) */ + sum0 += *pIn1++ * (*pIn2); + sum1 += *pIn1B++ * (*pIn2); + sum2 += *pIn1C++ * (*pIn2); + sum3 += *pIn1D++ * (*pIn2); + sum4 += *pIn1E++ * (*pIn2); + sum5 += *pIn1F++ * (*pIn2); + sum6 += *pIn1G++ * (*pIn2); + sum7 += *pIn1H++ * (*pIn2); + pIn2 += numColsB; + + /* Decrement the loop counter */ + colCnt--; + } + + /* Store the result in the destination buffer */ + *px++ = sum0; + *pxB++ = sum1; + *pxC++ = sum2; + *pxD++ = sum3; + *pxE++ = sum4; + *pxF++ = sum5; + *pxG++ = sum6; + *pxH++ = sum7; + + /* 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); + + /* Update the pointer pInA to point to the starting address of the next row */ + i = i + numColsB; + pInA = pInA + GROUPOFROWS*numColsA; + + /* Decrement the row loop counter */ + rowCnt--; + } + + /* + + i was the index of a group of rows computed by previous loop. + Now i is the index of a row since below code is computing row per row + and no more group of row per group of rows. + + */ + + i = GROUPOFROWS*i; + rowCnt = row & 7; + + while(rowCnt > 0) { /* Output pointer is set to starting address of the row being processed */ px = pOut + i; @@ -117,7 +313,7 @@ arm_status arm_mat_mult_f32( j = 0U; - /* column loop */ + /* Column loop */ do { /* Set the variable sum, that acts as accumulator, to zero */ @@ -126,43 +322,43 @@ arm_status arm_mat_mult_f32( /* 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. */ + acc0 = vdupq_n_f32(0.0); + + /* Compute 4 MACs simultaneously. */ colCnt = numColsA >> 2U; - /* matrix multiplication */ + /* 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) */ - in3 = *pIn2; + /* c(m,n) = a(1,1)*b(1,1) + a(1,2)*b(2,1) + ... + a(m,p)*b(p,n) */ + a0V = vld1q_f32(pIn1); // load & separate real/imag pSrcA (de-interleave 2) + pIn1 += 4; + + temp[0] = *pIn2; pIn2 += numColsB; - in1 = pIn1[0]; - in2 = pIn1[1]; - sum += in1 * in3; - in4 = *pIn2; + temp[1] = *pIn2; pIn2 += numColsB; - sum += in2 * in4; - - in3 = *pIn2; + temp[2] = *pIn2; pIn2 += numColsB; - in1 = pIn1[2]; - in2 = pIn1[3]; - sum += in1 * in3; - in4 = *pIn2; + temp[3] = *pIn2; pIn2 += numColsB; - sum += in2 * in4; - pIn1 += 4U; + + acc0 = vmlaq_f32(acc0,a0V,temp); /* Decrement the loop count */ colCnt--; } + accum = vpadd_f32(vget_low_f32(acc0), vget_high_f32(acc0)); + sum += accum[0] + accum[1]; + /* 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) */ + /* 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; @@ -182,40 +378,67 @@ arm_status arm_mat_mult_f32( } while (col > 0U); -#else - /* Run the below code for Cortex-M0 */ + /* Update the pointer pInA to point to the starting address of the next row */ + i = i + numColsB; + pInA = pInA + numColsA; - 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 */ + /* Decrement the row loop counter */ + rowCnt--; + + } + /* Set status as ARM_MATH_SUCCESS */ + status = ARM_MATH_SUCCESS; + } + + /* Return to application */ + return (status); +} +#else +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 *pInB = pSrcB->pData; /* Input data matrix pointer B */ + 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 */ + uint32_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)) + (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 */ + +#endif /* #ifdef ARM_MATH_MATRIX_CHECK */ { - /* The following loop performs the dot-product of each row in pInA with each column in pInB */ + /* 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 */ + /* Output pointer is set to starting address of row being processed */ px = pOut + i; - /* For every row wise process, the column loop counter is to be initiated */ + /* For every row wise process, 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 */ + /* For every row wise process, pIn2 pointer is set to starting address of pSrcB data */ pIn2 = pSrcB->pData; /* column loop */ @@ -224,43 +447,78 @@ arm_status arm_mat_mult_f32( /* 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 */ + /* Initialize pointer pIn1 to point to starting address of column being processed */ pIn1 = pInA; - /* Matrix A columns number of MAC operations are to be performed */ +#if defined (ARM_MATH_LOOPUNROLL) + + /* Loop unrolling: Compute 4 MACs at a time. */ + colCnt = numColsA >> 2U; + + /* 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) */ + + /* Perform the multiply-accumulates */ + sum += *pIn1++ * *pIn2; + pIn2 += numColsB; + + sum += *pIn1++ * *pIn2; + pIn2 += numColsB; + + sum += *pIn1++ * *pIn2; + pIn2 += numColsB; + + sum += *pIn1++ * *pIn2; + pIn2 += numColsB; + + /* Decrement loop counter */ + colCnt--; + } + + /* Loop unrolling: Compute remaining MACs */ + colCnt = numColsA % 0x4U; + +#else + + /* Initialize cntCnt with number of columns */ colCnt = numColsA; +#endif /* #if defined (ARM_MATH_LOOPUNROLL) */ + 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); + /* c(m,n) = a(1,1) * b(1,1) + a(1,2) * b(2,1) + .... + a(m,p) * b(p,n) */ + + /* Perform the multiply-accumulates */ + sum += *pIn1++ * *pIn2; pIn2 += numColsB; - /* Decrement the loop counter */ + /* Decrement loop counter */ colCnt--; } - /* Store the result in the destination buffer */ + /* Store result in destination buffer */ *px++ = sum; - /* Decrement the column loop counter */ + /* Decrement column loop counter */ col--; - /* Update the pointer pIn2 to point to the starting address of the next column */ + /* Update pointer pIn2 to point to starting address of next column */ pIn2 = pInB + (numColsB - col); } while (col > 0U); -#endif /* #if defined (ARM_MATH_DSP) */ - - /* Update the pointer pInA to point to the starting address of the next row */ + /* Update pointer pInA to point to starting address of next row */ i = i + numColsB; pInA = pInA + numColsA; - /* Decrement the row loop counter */ + /* Decrement row loop counter */ row--; } while (row > 0U); + /* Set status as ARM_MATH_SUCCESS */ status = ARM_MATH_SUCCESS; } @@ -269,6 +527,8 @@ arm_status arm_mat_mult_f32( return (status); } +#endif /* #if defined(ARM_MATH_NEON) */ + /** * @} end of MatrixMult group */ |