| 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891 |
- /* ----------------------------------------------------------------------
- * Project: CMSIS DSP Library
- * Title: arm_mat_inverse_f16.c
- * Description: Floating-point matrix inverse
- *
- * $Date: 23 April 2021
- * $Revision: V1.9.0
- *
- * Target Processor: Cortex-M and Cortex-A cores
- * -------------------------------------------------------------------- */
- /*
- * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved.
- *
- * SPDX-License-Identifier: Apache-2.0
- *
- * Licensed under the Apache License, Version 2.0 (the License); you may
- * not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an AS IS BASIS, WITHOUT
- * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
- #include "dsp/matrix_functions_f16.h"
- #if defined(ARM_FLOAT16_SUPPORTED)
- /**
- @ingroup groupMatrix
- */
- /**
- @addtogroup MatrixInv
- @{
- */
- /**
- @brief Floating-point matrix inverse.
- @param[in] pSrc points to input matrix structure. The source matrix is modified by the function.
- @param[out] pDst points to output matrix structure
- @return execution status
- - \ref ARM_MATH_SUCCESS : Operation successful
- - \ref ARM_MATH_SIZE_MISMATCH : Matrix size check failed
- - \ref ARM_MATH_SINGULAR : Input matrix is found to be singular (non-invertible)
- */
- #if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
- arm_status arm_mat_inverse_f16(
- const arm_matrix_instance_f16 * pSrc,
- arm_matrix_instance_f16 * pDst)
- {
- float16_t *pIn = pSrc->pData; /* input data matrix pointer */
- float16_t *pOut = pDst->pData; /* output data matrix pointer */
- float16_t *pInT1, *pInT2; /* Temporary input data matrix pointer */
- float16_t *pOutT1, *pOutT2; /* Temporary output data matrix pointer */
- float16_t *pPivotRowIn, *pPRT_in, *pPivotRowDst, *pPRT_pDst; /* Temporary input and output data matrix pointer */
- uint32_t numRows = pSrc->numRows; /* Number of rows in the matrix */
- uint32_t numCols = pSrc->numCols; /* Number of Cols in the matrix */
- float16_t *pTmpA, *pTmpB;
- _Float16 in = 0.0f16; /* Temporary input values */
- uint32_t i, rowCnt, flag = 0U, j, loopCnt, l; /* loop counters */
- arm_status status; /* status of matrix inverse */
- uint32_t blkCnt;
- #ifdef ARM_MATH_MATRIX_CHECK
- /* Check for matrix mismatch condition */
- if ((pSrc->numRows != pSrc->numCols) || (pDst->numRows != pDst->numCols)
- || (pSrc->numRows != pDst->numRows))
- {
- /* Set status as ARM_MATH_SIZE_MISMATCH */
- status = ARM_MATH_SIZE_MISMATCH;
- }
- else
- #endif /* #ifdef ARM_MATH_MATRIX_CHECK */
- {
- /*--------------------------------------------------------------------------------------------------------------
- * Matrix Inverse can be solved using elementary row operations.
- *
- * Gauss-Jordan Method:
- *
- * 1. First combine the identity matrix and the input matrix separated by a bar to form an
- * augmented matrix as follows:
- * _ _ _ _ _ _ _ _
- * | | a11 a12 | | | 1 0 | | | X11 X12 |
- * | | | | | | | = | |
- * |_ |_ a21 a22 _| | |_0 1 _| _| |_ X21 X21 _|
- *
- * 2. In our implementation, pDst Matrix is used as identity matrix.
- *
- * 3. Begin with the first row. Let i = 1.
- *
- * 4. Check to see if the pivot for row i is zero.
- * The pivot is the element of the main diagonal that is on the current row.
- * For instance, if working with row i, then the pivot element is aii.
- * If the pivot is zero, exchange that row with a row below it that does not
- * contain a zero in column i. If this is not possible, then an inverse
- * to that matrix does not exist.
- *
- * 5. Divide every element of row i by the pivot.
- *
- * 6. For every row below and row i, replace that row with the sum of that row and
- * a multiple of row i so that each new element in column i below row i is zero.
- *
- * 7. Move to the next row and column and repeat steps 2 through 5 until you have zeros
- * for every element below and above the main diagonal.
- *
- * 8. Now an identical matrix is formed to the left of the bar(input matrix, src).
- * Therefore, the matrix to the right of the bar is our solution(dst matrix, dst).
- *----------------------------------------------------------------------------------------------------------------*/
- /*
- * Working pointer for destination matrix
- */
- pOutT1 = pOut;
- /*
- * Loop over the number of rows
- */
- rowCnt = numRows;
- /*
- * Making the destination matrix as identity matrix
- */
- while (rowCnt > 0U)
- {
- /*
- * Writing all zeroes in lower triangle of the destination matrix
- */
- j = numRows - rowCnt;
- while (j > 0U)
- {
- *pOutT1++ = 0.0f16;
- j--;
- }
- /*
- * Writing all ones in the diagonal of the destination matrix
- */
- *pOutT1++ = 1.0f16;
- /*
- * Writing all zeroes in upper triangle of the destination matrix
- */
- j = rowCnt - 1U;
- while (j > 0U)
- {
- *pOutT1++ = 0.0f16;
- j--;
- }
- /*
- * Decrement the loop counter
- */
- rowCnt--;
- }
- /*
- * Loop over the number of columns of the input matrix.
- * All the elements in each column are processed by the row operations
- */
- loopCnt = numCols;
- /*
- * Index modifier to navigate through the columns
- */
- l = 0U;
- while (loopCnt > 0U)
- {
- /*
- * Check if the pivot element is zero..
- * If it is zero then interchange the row with non zero row below.
- * If there is no non zero element to replace in the rows below,
- * then the matrix is Singular.
- */
- /*
- * Working pointer for the input matrix that points
- * * to the pivot element of the particular row
- */
- pInT1 = pIn + (l * numCols);
- /*
- * Working pointer for the destination matrix that points
- * * to the pivot element of the particular row
- */
- pOutT1 = pOut + (l * numCols);
- /*
- * Temporary variable to hold the pivot value
- */
- in = *pInT1;
-
- /*
- * Check if the pivot element is zero
- */
- if ((_Float16)*pInT1 == 0.0f16)
- {
- /*
- * Loop over the number rows present below
- */
- for (i = 1U; i < numRows-l; i++)
- {
- /*
- * Update the input and destination pointers
- */
- pInT2 = pInT1 + (numCols * i);
- pOutT2 = pOutT1 + (numCols * i);
- /*
- * Check if there is a non zero pivot element to
- * * replace in the rows below
- */
- if ((_Float16)*pInT2 != 0.0f16)
- {
- f16x8_t vecA, vecB;
- /*
- * Loop over number of columns
- * * to the right of the pilot element
- */
- pTmpA = pInT1;
- pTmpB = pInT2;
- blkCnt = (numCols - l) >> 3;
- while (blkCnt > 0U)
- {
-
- vecA = vldrhq_f16(pTmpA);
- vecB = vldrhq_f16(pTmpB);
- vstrhq_f16(pTmpB, vecA);
- vstrhq_f16(pTmpA, vecB);
- pTmpA += 8;
- pTmpB += 8;
- /*
- * Decrement the blockSize loop counter
- */
- blkCnt--;
- }
- /*
- * tail
- * (will be merged thru tail predication)
- */
- blkCnt = (numCols - l) & 7;
- if (blkCnt > 0U)
- {
- mve_pred16_t p0 = vctp16q(blkCnt);
- vecA = vldrhq_f16(pTmpA);
- vecB = vldrhq_f16(pTmpB);
- vstrhq_p_f16(pTmpB, vecA, p0);
- vstrhq_p_f16(pTmpA, vecB, p0);
- }
- pInT1 += numCols - l;
- pInT2 += numCols - l;
- pTmpA = pOutT1;
- pTmpB = pOutT2;
- blkCnt = numCols >> 3;
- while (blkCnt > 0U)
- {
- vecA = vldrhq_f16(pTmpA);
- vecB = vldrhq_f16(pTmpB);
- vstrhq_f16(pTmpB, vecA);
- vstrhq_f16(pTmpA, vecB);
- pTmpA += 8;
- pTmpB += 8;
- /*
- * Decrement the blockSize loop counter
- */
- blkCnt--;
- }
- /*
- * tail
- */
- blkCnt = numCols & 7;
- if (blkCnt > 0U)
- {
- mve_pred16_t p0 = vctp16q(blkCnt);
- vecA = vldrhq_f16(pTmpA);
- vecB = vldrhq_f16(pTmpB);
- vstrhq_p_f16(pTmpB, vecA, p0);
- vstrhq_p_f16(pTmpA, vecB, p0);
- }
- pOutT1 += numCols;
- pOutT2 += numCols;
- /*
- * Flag to indicate whether exchange is done or not
- */
- flag = 1U;
- /*
- * Break after exchange is done
- */
- break;
- }
-
- }
- }
- /*
- * Update the status if the matrix is singular
- */
- if ((flag != 1U) && (in == 0.0f16))
- {
- return ARM_MATH_SINGULAR;
- }
- /*
- * Points to the pivot row of input and destination matrices
- */
- pPivotRowIn = pIn + (l * numCols);
- pPivotRowDst = pOut + (l * numCols);
- /*
- * Temporary pointers to the pivot row pointers
- */
- pInT1 = pPivotRowIn;
- pOutT1 = pPivotRowDst;
- /*
- * Pivot element of the row
- */
- in = *(pIn + (l * numCols));
- pTmpA = pInT1;
- f16x8_t invIn = vdupq_n_f16(1.0f16 / in);
- blkCnt = (numCols - l) >> 3;
- f16x8_t vecA;
- while (blkCnt > 0U)
- {
- *(f16x8_t *) pTmpA = *(f16x8_t *) pTmpA * invIn;
- pTmpA += 8;
- /*
- * Decrement the blockSize loop counter
- */
- blkCnt--;
- }
- /*
- * tail
- */
- blkCnt = (numCols - l) & 7;
- if (blkCnt > 0U)
- {
- mve_pred16_t p0 = vctp16q(blkCnt);
-
- vecA = vldrhq_f16(pTmpA);
- vecA = vecA * invIn;
- vstrhq_p_f16(pTmpA, vecA, p0);
- }
- pInT1 += numCols - l;
- /*
- * Loop over number of columns
- * * to the right of the pilot element
- */
- pTmpA = pOutT1;
- blkCnt = numCols >> 3;
- while (blkCnt > 0U)
- {
- *(f16x8_t *) pTmpA = *(f16x8_t *) pTmpA *invIn;
- pTmpA += 8;
- /*
- * Decrement the blockSize loop counter
- */
- blkCnt--;
- }
- /*
- * tail
- * (will be merged thru tail predication)
- */
- blkCnt = numCols & 7;
- if (blkCnt > 0U)
- {
- mve_pred16_t p0 = vctp16q(blkCnt);
- vecA = vldrhq_f16(pTmpA);
- vecA = vecA * invIn;
- vstrhq_p_f16(pTmpA, vecA, p0);
- }
- pOutT1 += numCols;
- /*
- * Replace the rows with the sum of that row and a multiple of row i
- * * so that each new element in column i above row i is zero.
- */
- /*
- * Temporary pointers for input and destination matrices
- */
- pInT1 = pIn;
- pOutT1 = pOut;
- for (i = 0U; i < numRows; i++)
- {
- /*
- * Check for the pivot element
- */
- if (i == l)
- {
- /*
- * If the processing element is the pivot element,
- * only the columns to the right are to be processed
- */
- pInT1 += numCols - l;
- pOutT1 += numCols;
- }
- else
- {
- /*
- * Element of the reference row
- */
- /*
- * Working pointers for input and destination pivot rows
- */
- pPRT_in = pPivotRowIn;
- pPRT_pDst = pPivotRowDst;
- /*
- * Loop over the number of columns to the right of the pivot element,
- * to replace the elements in the input matrix
- */
- in = *pInT1;
- f16x8_t tmpV = vdupq_n_f16(in);
- blkCnt = (numCols - l) >> 3;
- while (blkCnt > 0U)
- {
- f16x8_t vec1, vec2;
- /*
- * Replace the element by the sum of that row
- * and a multiple of the reference row
- */
- vec1 = vldrhq_f16(pInT1);
- vec2 = vldrhq_f16(pPRT_in);
- vec1 = vfmsq_f16(vec1, tmpV, vec2);
- vstrhq_f16(pInT1, vec1);
- pPRT_in += 8;
- pInT1 += 8;
- /*
- * Decrement the blockSize loop counter
- */
- blkCnt--;
- }
- /*
- * tail
- * (will be merged thru tail predication)
- */
- blkCnt = (numCols - l) & 7;
- if (blkCnt > 0U)
- {
- f16x8_t vec1, vec2;
- mve_pred16_t p0 = vctp16q(blkCnt);
- vec1 = vldrhq_f16(pInT1);
- vec2 = vldrhq_f16(pPRT_in);
- vec1 = vfmsq_f16(vec1, tmpV, vec2);
- vstrhq_p_f16(pInT1, vec1, p0);
- pInT1 += blkCnt;
- }
- blkCnt = numCols >> 3;
- while (blkCnt > 0U)
- {
- f16x8_t vec1, vec2;
- /*
- * Replace the element by the sum of that row
- * and a multiple of the reference row
- */
- vec1 = vldrhq_f16(pOutT1);
- vec2 = vldrhq_f16(pPRT_pDst);
- vec1 = vfmsq_f16(vec1, tmpV, vec2);
- vstrhq_f16(pOutT1, vec1);
- pPRT_pDst += 8;
- pOutT1 += 8;
- /*
- * Decrement the blockSize loop counter
- */
- blkCnt--;
- }
- /*
- * tail
- * (will be merged thru tail predication)
- */
- blkCnt = numCols & 7;
- if (blkCnt > 0U)
- {
- f16x8_t vec1, vec2;
- mve_pred16_t p0 = vctp16q(blkCnt);
- vec1 = vldrhq_f16(pOutT1);
- vec2 = vldrhq_f16(pPRT_pDst);
- vec1 = vfmsq_f16(vec1, tmpV, vec2);
- vstrhq_p_f16(pOutT1, vec1, p0);
- pInT2 += blkCnt;
- pOutT1 += blkCnt;
- }
- }
- /*
- * Increment the temporary input pointer
- */
- pInT1 = pInT1 + l;
- }
- /*
- * Increment the input pointer
- */
- pIn++;
- /*
- * Decrement the loop counter
- */
- loopCnt--;
- /*
- * Increment the index modifier
- */
- l++;
- }
- /*
- * Set status as ARM_MATH_SUCCESS
- */
- status = ARM_MATH_SUCCESS;
- if ((flag != 1U) && (in == 0.0f16))
- {
- pIn = pSrc->pData;
- for (i = 0; i < numRows * numCols; i++)
- {
- if ((_Float16)pIn[i] != 0.0f16)
- break;
- }
- if (i == numRows * numCols)
- status = ARM_MATH_SINGULAR;
- }
- }
- /* Return to application */
- return (status);
- }
- #else
- arm_status arm_mat_inverse_f16(
- const arm_matrix_instance_f16 * pSrc,
- arm_matrix_instance_f16 * pDst)
- {
- float16_t *pIn = pSrc->pData; /* input data matrix pointer */
- float16_t *pOut = pDst->pData; /* output data matrix pointer */
- float16_t *pInT1, *pInT2; /* Temporary input data matrix pointer */
- float16_t *pOutT1, *pOutT2; /* Temporary output data matrix pointer */
- float16_t *pPivotRowIn, *pPRT_in, *pPivotRowDst, *pPRT_pDst; /* Temporary input and output data matrix pointer */
- uint32_t numRows = pSrc->numRows; /* Number of rows in the matrix */
- uint32_t numCols = pSrc->numCols; /* Number of Cols in the matrix */
- _Float16 Xchg, in = 0.0f16, in1; /* Temporary input values */
- uint32_t i, rowCnt, flag = 0U, j, loopCnt, k,l; /* loop counters */
- arm_status status; /* status of matrix inverse */
- #ifdef ARM_MATH_MATRIX_CHECK
- /* Check for matrix mismatch condition */
- if ((pSrc->numRows != pSrc->numCols) ||
- (pDst->numRows != pDst->numCols) ||
- (pSrc->numRows != pDst->numRows) )
- {
- /* Set status as ARM_MATH_SIZE_MISMATCH */
- status = ARM_MATH_SIZE_MISMATCH;
- }
- else
- #endif /* #ifdef ARM_MATH_MATRIX_CHECK */
- {
- /*--------------------------------------------------------------------------------------------------------------
- * Matrix Inverse can be solved using elementary row operations.
- *
- * Gauss-Jordan Method:
- *
- * 1. First combine the identity matrix and the input matrix separated by a bar to form an
- * augmented matrix as follows:
- * _ _ _ _
- * | a11 a12 | 1 0 | | X11 X12 |
- * | | | = | |
- * |_ a21 a22 | 0 1 _| |_ X21 X21 _|
- *
- * 2. In our implementation, pDst Matrix is used as identity matrix.
- *
- * 3. Begin with the first row. Let i = 1.
- *
- * 4. Check to see if the pivot for row i is zero.
- * The pivot is the element of the main diagonal that is on the current row.
- * For instance, if working with row i, then the pivot element is aii.
- * If the pivot is zero, exchange that row with a row below it that does not
- * contain a zero in column i. If this is not possible, then an inverse
- * to that matrix does not exist.
- *
- * 5. Divide every element of row i by the pivot.
- *
- * 6. For every row below and row i, replace that row with the sum of that row and
- * a multiple of row i so that each new element in column i below row i is zero.
- *
- * 7. Move to the next row and column and repeat steps 2 through 5 until you have zeros
- * for every element below and above the main diagonal.
- *
- * 8. Now an identical matrix is formed to the left of the bar(input matrix, pSrc).
- * Therefore, the matrix to the right of the bar is our solution(pDst matrix, pDst).
- *----------------------------------------------------------------------------------------------------------------*/
- /* Working pointer for destination matrix */
- pOutT1 = pOut;
- /* Loop over the number of rows */
- rowCnt = numRows;
- /* Making the destination matrix as identity matrix */
- while (rowCnt > 0U)
- {
- /* Writing all zeroes in lower triangle of the destination matrix */
- j = numRows - rowCnt;
- while (j > 0U)
- {
- *pOutT1++ = 0.0f16;
- j--;
- }
- /* Writing all ones in the diagonal of the destination matrix */
- *pOutT1++ = 1.0f16;
- /* Writing all zeroes in upper triangle of the destination matrix */
- j = rowCnt - 1U;
- while (j > 0U)
- {
- *pOutT1++ = 0.0f16;
- j--;
- }
- /* Decrement loop counter */
- rowCnt--;
- }
- /* Loop over the number of columns of the input matrix.
- All the elements in each column are processed by the row operations */
- loopCnt = numCols;
- /* Index modifier to navigate through the columns */
- l = 0U;
- while (loopCnt > 0U)
- {
- /* Check if the pivot element is zero..
- * If it is zero then interchange the row with non zero row below.
- * If there is no non zero element to replace in the rows below,
- * then the matrix is Singular. */
- /* Working pointer for the input matrix that points
- * to the pivot element of the particular row */
- pInT1 = pIn + (l * numCols);
- /* Working pointer for the destination matrix that points
- * to the pivot element of the particular row */
- pOutT1 = pOut + (l * numCols);
- /* Temporary variable to hold the pivot value */
- in = *pInT1;
- /* Check if the pivot element is zero */
- if ((_Float16)*pInT1 == 0.0f16)
- {
- /* Loop over the number rows present below */
- for (i = 1U; i < numRows-l; i++)
- {
- /* Update the input and destination pointers */
- pInT2 = pInT1 + (numCols * i);
- pOutT2 = pOutT1 + (numCols * i);
- /* Check if there is a non zero pivot element to
- * replace in the rows below */
- if ((_Float16)*pInT2 != 0.0f16)
- {
- /* Loop over number of columns
- * to the right of the pilot element */
- j = numCols - l;
- while (j > 0U)
- {
- /* Exchange the row elements of the input matrix */
- Xchg = *pInT2;
- *pInT2++ = *pInT1;
- *pInT1++ = Xchg;
- /* Decrement the loop counter */
- j--;
- }
- /* Loop over number of columns of the destination matrix */
- j = numCols;
- while (j > 0U)
- {
- /* Exchange the row elements of the destination matrix */
- Xchg = *pOutT2;
- *pOutT2++ = *pOutT1;
- *pOutT1++ = Xchg;
- /* Decrement loop counter */
- j--;
- }
- /* Flag to indicate whether exchange is done or not */
- flag = 1U;
- /* Break after exchange is done */
- break;
- }
- }
- }
- /* Update the status if the matrix is singular */
- if ((flag != 1U) && (in == 0.0f16))
- {
- return ARM_MATH_SINGULAR;
- }
- /* Points to the pivot row of input and destination matrices */
- pPivotRowIn = pIn + (l * numCols);
- pPivotRowDst = pOut + (l * numCols);
- /* Temporary pointers to the pivot row pointers */
- pInT1 = pPivotRowIn;
- pInT2 = pPivotRowDst;
- /* Pivot element of the row */
- in = *pPivotRowIn;
- /* Loop over number of columns
- * to the right of the pilot element */
- j = (numCols - l);
- while (j > 0U)
- {
- /* Divide each element of the row of the input matrix
- * by the pivot element */
- in1 = *pInT1;
- *pInT1++ = in1 / in;
- /* Decrement the loop counter */
- j--;
- }
- /* Loop over number of columns of the destination matrix */
- j = numCols;
- while (j > 0U)
- {
- /* Divide each element of the row of the destination matrix
- * by the pivot element */
- in1 = *pInT2;
- *pInT2++ = in1 / in;
- /* Decrement the loop counter */
- j--;
- }
- /* Replace the rows with the sum of that row and a multiple of row i
- * so that each new element in column i above row i is zero.*/
- /* Temporary pointers for input and destination matrices */
- pInT1 = pIn;
- pInT2 = pOut;
- /* index used to check for pivot element */
- i = 0U;
- /* Loop over number of rows */
- /* to be replaced by the sum of that row and a multiple of row i */
- k = numRows;
- while (k > 0U)
- {
- /* Check for the pivot element */
- if (i == l)
- {
- /* If the processing element is the pivot element,
- only the columns to the right are to be processed */
- pInT1 += numCols - l;
- pInT2 += numCols;
- }
- else
- {
- /* Element of the reference row */
- in = *pInT1;
- /* Working pointers for input and destination pivot rows */
- pPRT_in = pPivotRowIn;
- pPRT_pDst = pPivotRowDst;
- /* Loop over the number of columns to the right of the pivot element,
- to replace the elements in the input matrix */
- j = (numCols - l);
- while (j > 0U)
- {
- /* Replace the element by the sum of that row
- and a multiple of the reference row */
- in1 = *pInT1;
- *pInT1++ = (_Float16)in1 - ((_Float16)in * (_Float16)*pPRT_in++);
- /* Decrement the loop counter */
- j--;
- }
- /* Loop over the number of columns to
- replace the elements in the destination matrix */
- j = numCols;
- while (j > 0U)
- {
- /* Replace the element by the sum of that row
- and a multiple of the reference row */
- in1 = *pInT2;
- *pInT2++ = (_Float16)in1 - ((_Float16)in * (_Float16)*pPRT_pDst++);
- /* Decrement loop counter */
- j--;
- }
- }
- /* Increment temporary input pointer */
- pInT1 = pInT1 + l;
- /* Decrement loop counter */
- k--;
- /* Increment pivot index */
- i++;
- }
- /* Increment the input pointer */
- pIn++;
- /* Decrement the loop counter */
- loopCnt--;
- /* Increment the index modifier */
- l++;
- }
- /* Set status as ARM_MATH_SUCCESS */
- status = ARM_MATH_SUCCESS;
- if ((flag != 1U) && ((_Float16)in == 0.0f16))
- {
- pIn = pSrc->pData;
- for (i = 0; i < numRows * numCols; i++)
- {
- if ((_Float16)pIn[i] != 0.0f16)
- break;
- }
- if (i == numRows * numCols)
- status = ARM_MATH_SINGULAR;
- }
- }
- /* Return to application */
- return (status);
- }
- #endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */
- /**
- @} end of MatrixInv group
- */
- #endif /* #if defined(ARM_FLOAT16_SUPPORTED) */
|