diff options
Diffstat (limited to 'Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_conv_q7.c')
| -rw-r--r-- | Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_conv_q7.c | 302 |
1 files changed, 162 insertions, 140 deletions
diff --git a/Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_conv_q7.c b/Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_conv_q7.c index 9e5a79b58..bdd1cab89 100644 --- a/Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_conv_q7.c +++ b/Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_conv_q7.c @@ -3,13 +3,13 @@ * Title: arm_conv_q7.c * Description: Convolution of Q7 sequences * - * $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 * @@ -29,62 +29,60 @@ #include "arm_math.h" /** - * @ingroup groupFilters + @ingroup groupFilters */ /** - * @addtogroup Conv - * @{ + @addtogroup Conv + @{ */ /** - * @brief Convolution of Q7 sequences. - * @param[in] *pSrcA points to the first input sequence. - * @param[in] srcALen length of the first input sequence. - * @param[in] *pSrcB points to the second input sequence. - * @param[in] srcBLen length of the second input sequence. - * @param[out] *pDst points to the location where the output result is written. Length srcALen+srcBLen-1. - * @return none. - * - * @details - * <b>Scaling and Overflow Behavior:</b> - * - * \par - * The function is implemented using a 32-bit internal accumulator. - * Both the inputs are represented in 1.7 format and multiplications yield a 2.14 result. - * The 2.14 intermediate results are accumulated in a 32-bit accumulator in 18.14 format. - * This approach provides 17 guard bits and there is no risk of overflow as long as <code>max(srcALen, srcBLen)<131072</code>. - * The 18.14 result is then truncated to 18.7 format by discarding the low 7 bits and then saturated to 1.7 format. - * - * \par - * Refer the function <code>arm_conv_opt_q7()</code> for a faster implementation of this function. - * + @brief Convolution of Q7 sequences. + @param[in] pSrcA points to the first input sequence + @param[in] srcALen length of the first input sequence + @param[in] pSrcB points to the second input sequence + @param[in] srcBLen length of the second input sequence + @param[out] pDst points to the location where the output result is written. Length srcALen+srcBLen-1. + @return none + + @par Scaling and Overflow Behavior + The function is implemented using a 32-bit internal accumulator. + Both the inputs are represented in 1.7 format and multiplications yield a 2.14 result. + The 2.14 intermediate results are accumulated in a 32-bit accumulator in 18.14 format. + This approach provides 17 guard bits and there is no risk of overflow as long as <code>max(srcALen, srcBLen)<131072</code>. + The 18.14 result is then truncated to 18.7 format by discarding the low 7 bits and then saturated to 1.7 format. + @remark + Refer to \ref arm_conv_opt_q7() for a faster implementation of this function. */ void arm_conv_q7( - q7_t * pSrcA, - uint32_t srcALen, - q7_t * pSrcB, - uint32_t srcBLen, - q7_t * pDst) + const q7_t * pSrcA, + uint32_t srcALen, + const q7_t * pSrcB, + uint32_t srcBLen, + q7_t * pDst) { - -#if defined (ARM_MATH_DSP) - - /* Run the below code for Cortex-M4 and Cortex-M3 */ - - q7_t *pIn1; /* inputA pointer */ - q7_t *pIn2; /* inputB pointer */ - q7_t *pOut = pDst; /* output pointer */ - q7_t *px; /* Intermediate inputA pointer */ - q7_t *py; /* Intermediate inputB pointer */ - q7_t *pSrc1, *pSrc2; /* Intermediate pointers */ - q7_t x0, x1, x2, x3, c0, c1; /* Temporary variables to hold state and coefficient values */ - q31_t sum, acc0, acc1, acc2, acc3; /* Accumulator */ - q31_t input1, input2; /* Temporary input variables */ - q15_t in1, in2; /* Temporary input variables */ - uint32_t j, k, count, blkCnt, blockSize1, blockSize2, blockSize3; /* loop counter */ +#if (1) +//#if !defined(ARM_MATH_CM0_FAMILY) + + const q7_t *pIn1; /* InputA pointer */ + const q7_t *pIn2; /* InputB pointer */ + q7_t *pOut = pDst; /* Output pointer */ + const q7_t *px; /* Intermediate inputA pointer */ + const q7_t *py; /* Intermediate inputB pointer */ + const q7_t *pSrc1, *pSrc2; /* Intermediate pointers */ + q31_t sum; /* Accumulators */ + uint32_t blockSize1, blockSize2, blockSize3; /* Loop counters */ + uint32_t j, k, count, blkCnt; /* Loop counters */ + +#if defined (ARM_MATH_LOOPUNROLL) + q31_t acc0, acc1, acc2, acc3; /* Accumulators */ + q31_t input1, input2; /* Temporary input variables */ + q15_t in1, in2; /* Temporary input variables */ + q7_t x0, x1, x2, x3, c0, c1; /* Temporary variables to hold state and coefficient values */ +#endif /* The algorithm implementation is based on the lengths of the inputs. */ /* srcB is always made to slide across srcA. */ @@ -123,7 +121,7 @@ void arm_conv_q7( /* The algorithm is implemented in three stages. The loop counters of each stage is initiated here. */ blockSize1 = srcBLen - 1U; - blockSize2 = (srcALen - srcBLen) + 1U; + blockSize2 = srcALen - (srcBLen - 1U); blockSize3 = blockSize1; /* -------------------------- @@ -157,21 +155,21 @@ void arm_conv_q7( /* Accumulator is made zero for every iteration */ sum = 0; - /* Apply loop unrolling and compute 4 MACs simultaneously. */ +#if defined (ARM_MATH_LOOPUNROLL) + + /* Loop unrolling: Compute 4 outputs at a time */ k = count >> 2U; - /* First part of the processing with loop unrolling. Compute 4 MACs at a time. - ** a second loop below computes MACs for the remaining 1 to 3 samples. */ while (k > 0U) { /* x[0] , x[1] */ - in1 = (q15_t) * px++; - in2 = (q15_t) * px++; + in1 = (q15_t) *px++; + in2 = (q15_t) *px++; input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); /* y[srcBLen - 1] , y[srcBLen - 2] */ - in1 = (q15_t) * py--; - in2 = (q15_t) * py--; + in1 = (q15_t) *py--; + in2 = (q15_t) *py--; input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); /* x[0] * y[srcBLen - 1] */ @@ -179,33 +177,39 @@ void arm_conv_q7( sum = __SMLAD(input1, input2, sum); /* x[2] , x[3] */ - in1 = (q15_t) * px++; - in2 = (q15_t) * px++; + in1 = (q15_t) *px++; + in2 = (q15_t) *px++; input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); /* y[srcBLen - 3] , y[srcBLen - 4] */ - in1 = (q15_t) * py--; - in2 = (q15_t) * py--; + in1 = (q15_t) *py--; + in2 = (q15_t) *py--; input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); /* x[2] * y[srcBLen - 3] */ /* x[3] * y[srcBLen - 4] */ sum = __SMLAD(input1, input2, sum); - /* Decrement the loop counter */ + /* Decrement loop counter */ k--; } - /* If the count is not a multiple of 4, compute any remaining MACs here. - ** No loop unrolling is used. */ + /* Loop unrolling: Compute remaining outputs */ k = count % 0x4U; +#else + + /* Initialize k with number of samples */ + k = count; + +#endif /* #if defined (ARM_MATH_LOOPUNROLL) */ + while (k > 0U) { - /* Perform the multiply-accumulates */ - sum += ((q15_t) * px++ * *py--); + /* Perform the multiply-accumulate */ + sum += ((q15_t) *px++ * *py--); - /* Decrement the loop counter */ + /* Decrement loop counter */ k--; } @@ -216,10 +220,10 @@ void arm_conv_q7( py = pIn2 + count; px = pIn1; - /* Increment the MAC count */ + /* Increment MAC count */ count++; - /* Decrement the loop counter */ + /* Decrement loop counter */ blockSize1--; } @@ -228,7 +232,7 @@ void arm_conv_q7( * ------------------------*/ /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0] - * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0] + * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0] * .... * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0] */ @@ -252,7 +256,9 @@ void arm_conv_q7( * srcBLen should be greater than or equal to 4 */ if (srcBLen >= 4U) { - /* Loop unroll over blockSize2, by 4 */ +#if defined (ARM_MATH_LOOPUNROLL) + + /* Loop unrolling: Compute 4 outputs at a time */ blkCnt = blockSize2 >> 2U; while (blkCnt > 0U) @@ -264,9 +270,9 @@ void arm_conv_q7( acc3 = 0; /* read x[0], x[1], x[2] samples */ - x0 = *(px++); - x1 = *(px++); - x2 = *(px++); + x0 = *px++; + x1 = *px++; + x2 = *px++; /* Apply loop unrolling and compute 4 MACs simultaneously. */ k = srcBLen >> 2U; @@ -276,12 +282,12 @@ void arm_conv_q7( do { /* Read y[srcBLen - 1] sample */ - c0 = *(py--); + c0 = *py--; /* Read y[srcBLen - 2] sample */ - c1 = *(py--); + c1 = *py--; /* Read x[3] sample */ - x3 = *(px++); + x3 = *px++; /* x[0] and x[1] are packed */ in1 = (q15_t) x0; @@ -317,7 +323,7 @@ void arm_conv_q7( acc2 = __SMLAD(input1, input2, acc2); /* Read x[4] sample */ - x0 = *(px++); + x0 = *px++; /* x[3] and x[4] are packed */ in1 = (q15_t) x3; @@ -329,12 +335,12 @@ void arm_conv_q7( acc3 = __SMLAD(input1, input2, acc3); /* Read y[srcBLen - 3] sample */ - c0 = *(py--); + c0 = *py--; /* Read y[srcBLen - 4] sample */ - c1 = *(py--); + c1 = *py--; /* Read x[5] sample */ - x1 = *(px++); + x1 = *px++; /* x[2] and x[3] are packed */ in1 = (q15_t) x2; @@ -370,7 +376,7 @@ void arm_conv_q7( acc2 = __SMLAD(input1, input2, acc2); /* Read x[6] sample */ - x2 = *(px++); + x2 = *px++; /* x[5] and x[6] are packed */ in1 = (q15_t) x1; @@ -390,10 +396,9 @@ void arm_conv_q7( while (k > 0U) { /* Read y[srcBLen - 5] sample */ - c0 = *(py--); - + c0 = *py--; /* Read x[7] sample */ - x3 = *(px++); + x3 = *px++; /* Perform the multiply-accumulates */ /* acc0 += x[4] * y[srcBLen - 5] */ @@ -410,11 +415,10 @@ void arm_conv_q7( x1 = x2; x2 = x3; - /* Decrement the loop counter */ + /* Decrement loop counter */ k--; } - /* Store the result in the accumulator in the destination buffer. */ *pOut++ = (q7_t) (__SSAT(acc0 >> 7U, 8)); *pOut++ = (q7_t) (__SSAT(acc1 >> 7U, 8)); @@ -428,65 +432,77 @@ void arm_conv_q7( px = pIn1 + count; py = pSrc2; - /* Decrement the loop counter */ + /* Decrement loop counter */ blkCnt--; } - /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here. - ** No loop unrolling is used. */ + /* Loop unrolling: Compute remaining outputs */ blkCnt = blockSize2 % 0x4U; +#else + + /* Initialize blkCnt with number of samples */ + blkCnt = blockSize2; + +#endif /* #if defined (ARM_MATH_LOOPUNROLL) */ + while (blkCnt > 0U) { /* Accumulator is made zero for every iteration */ sum = 0; - /* Apply loop unrolling and compute 4 MACs simultaneously. */ +#if defined (ARM_MATH_LOOPUNROLL) + + /* Loop unrolling: Compute 4 outputs at a time */ k = srcBLen >> 2U; - /* First part of the processing with loop unrolling. Compute 4 MACs at a time. - ** a second loop below computes MACs for the remaining 1 to 3 samples. */ while (k > 0U) { /* Reading two inputs of SrcA buffer and packing */ - in1 = (q15_t) * px++; - in2 = (q15_t) * px++; + in1 = (q15_t) *px++; + in2 = (q15_t) *px++; input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); /* Reading two inputs of SrcB buffer and packing */ - in1 = (q15_t) * py--; - in2 = (q15_t) * py--; + in1 = (q15_t) *py--; + in2 = (q15_t) *py--; input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); - /* Perform the multiply-accumulates */ + /* Perform the multiply-accumulate */ sum = __SMLAD(input1, input2, sum); /* Reading two inputs of SrcA buffer and packing */ - in1 = (q15_t) * px++; - in2 = (q15_t) * px++; + in1 = (q15_t) *px++; + in2 = (q15_t) *px++; input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); /* Reading two inputs of SrcB buffer and packing */ - in1 = (q15_t) * py--; - in2 = (q15_t) * py--; + in1 = (q15_t) *py--; + in2 = (q15_t) *py--; input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); - /* Perform the multiply-accumulates */ + /* Perform the multiply-accumulate */ sum = __SMLAD(input1, input2, sum); - /* Decrement the loop counter */ + /* Decrement loop counter */ k--; } - /* If the srcBLen is not a multiple of 4, compute any remaining MACs here. - ** No loop unrolling is used. */ + /* Loop unrolling: Compute remaining outputs */ k = srcBLen % 0x4U; +#else + + /* Initialize blkCnt with number of samples */ + k = srcBLen; + +#endif /* #if defined (ARM_MATH_LOOPUNROLL) */ + while (k > 0U) { - /* Perform the multiply-accumulates */ - sum += ((q15_t) * px++ * *py--); + /* Perform the multiply-accumulate */ + sum += ((q15_t) *px++ * *py--); /* Decrement the loop counter */ k--; @@ -523,7 +539,7 @@ void arm_conv_q7( while (k > 0U) { /* Perform the multiply-accumulate */ - sum += ((q15_t) * px++ * *py--); + sum += ((q15_t) *px++ * *py--); /* Decrement the loop counter */ k--; @@ -539,7 +555,7 @@ void arm_conv_q7( px = pIn1 + count; py = pSrc2; - /* Decrement the loop counter */ + /* Decrement loop counter */ blkCnt--; } } @@ -576,21 +592,21 @@ void arm_conv_q7( /* Accumulator is made zero for every iteration */ sum = 0; - /* Apply loop unrolling and compute 4 MACs simultaneously. */ +#if defined (ARM_MATH_LOOPUNROLL) + + /* Loop unrolling: Compute 4 outputs at a time */ k = blockSize3 >> 2U; - /* First part of the processing with loop unrolling. Compute 4 MACs at a time. - ** a second loop below computes MACs for the remaining 1 to 3 samples. */ while (k > 0U) { /* Reading two inputs, x[srcALen - srcBLen + 1] and x[srcALen - srcBLen + 2] of SrcA buffer and packing */ - in1 = (q15_t) * px++; - in2 = (q15_t) * px++; + in1 = (q15_t) *px++; + in2 = (q15_t) *px++; input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); /* Reading two inputs, y[srcBLen - 1] and y[srcBLen - 2] of SrcB buffer and packing */ - in1 = (q15_t) * py--; - in2 = (q15_t) * py--; + in1 = (q15_t) *py--; + in2 = (q15_t) *py--; input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); /* sum += x[srcALen - srcBLen + 1] * y[srcBLen - 1] */ @@ -598,33 +614,40 @@ void arm_conv_q7( sum = __SMLAD(input1, input2, sum); /* Reading two inputs, x[srcALen - srcBLen + 3] and x[srcALen - srcBLen + 4] of SrcA buffer and packing */ - in1 = (q15_t) * px++; - in2 = (q15_t) * px++; + in1 = (q15_t) *px++; + in2 = (q15_t) *px++; input1 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); /* Reading two inputs, y[srcBLen - 3] and y[srcBLen - 4] of SrcB buffer and packing */ - in1 = (q15_t) * py--; - in2 = (q15_t) * py--; + in1 = (q15_t) *py--; + in2 = (q15_t) *py--; input2 = ((q31_t) in1 & 0x0000FFFF) | ((q31_t) in2 << 16U); /* sum += x[srcALen - srcBLen + 3] * y[srcBLen - 3] */ /* sum += x[srcALen - srcBLen + 4] * y[srcBLen - 4] */ sum = __SMLAD(input1, input2, sum); - /* Decrement the loop counter */ + /* Decrement loop counter */ k--; } - /* If the blockSize3 is not a multiple of 4, compute any remaining MACs here. - ** No loop unrolling is used. */ + /* Loop unrolling: Compute remaining outputs */ k = blockSize3 % 0x4U; +#else + + /* Initialize blkCnt with number of samples */ + k = blockSize3; + +#endif /* #if defined (ARM_MATH_LOOPUNROLL) */ + while (k > 0U) { - /* Perform the multiply-accumulates */ - sum += ((q15_t) * px++ * *py--); + /* Perform the multiply-accumulate */ + /* sum += x[srcALen-1] * y[srcBLen-1] */ + sum += ((q15_t) *px++ * *py--); - /* Decrement the loop counter */ + /* Decrement loop counter */ k--; } @@ -635,33 +658,32 @@ void arm_conv_q7( px = ++pSrc1; py = pSrc2; - /* Decrement the loop counter */ + /* Decrement loop counter */ blockSize3--; } #else +/* alternate version for CM0_FAMILY */ - /* Run the below code for Cortex-M0 */ - - q7_t *pIn1 = pSrcA; /* input pointer */ - q7_t *pIn2 = pSrcB; /* coefficient pointer */ - q31_t sum; /* Accumulator */ - uint32_t i, j; /* loop counter */ + const q7_t *pIn1 = pSrcA; /* InputA pointer */ + const q7_t *pIn2 = pSrcB; /* InputB pointer */ + q31_t sum; /* Accumulator */ + uint32_t i, j; /* Loop counters */ - /* Loop to calculate output of convolution for output length number of times */ - for (i = 0; i < (srcALen + srcBLen - 1); i++) + /* Loop to calculate convolution for output length number of times */ + for (i = 0U; i < (srcALen + srcBLen - 1U); i++) { - /* Initialize sum with zero to carry on MAC operations */ + /* Initialize sum with zero to carry out MAC operations */ sum = 0; /* Loop to perform MAC operations according to convolution equation */ - for (j = 0; j <= i; j++) + for (j = 0U; j <= i; j++) { /* Check the array limitations */ if (((i - j) < srcBLen) && (j < srcALen)) { /* z[i] += x[i-j] * y[j] */ - sum += (q15_t) pIn1[j] * (pIn2[i - j]); + sum += ((q15_t) pIn1[j] * pIn2[i - j]); } } @@ -669,10 +691,10 @@ void arm_conv_q7( pDst[i] = (q7_t) __SSAT((sum >> 7U), 8U); } -#endif /* #if defined (ARM_MATH_DSP) */ +#endif /* #if !defined(ARM_MATH_CM0_FAMILY) */ } /** - * @} end of Conv group + @} end of Conv group */ |