![](/media/cache/group/default_image.jpg.50x50_q85.jpg)
Revision 4:a89e836d9faf, committed 2017-05-12
- Comitter:
- elt14lpo
- Date:
- Fri May 12 08:01:26 2017 +0000
- Parent:
- 3:3f71950ceb71
- Commit message:
Changed in this revision
diff -r 3f71950ceb71 -r a89e836d9faf dsp_test.cpp --- a/dsp_test.cpp Fri May 12 06:59:55 2017 +0000 +++ b/dsp_test.cpp Fri May 12 08:01:26 2017 +0000 @@ -1,6 +1,6 @@ #include "mbed.h" #include <iostream> /* cout */ -//#include <stdio.h> /* printf */ +#include <stdio.h> /* printf */ #include <math.h> /* sin */ #include <vector> #include <stdlib.h> /* abs */ @@ -10,8 +10,6 @@ #include <DHT.h> #include<sstream> #include "luke_correlate_f32.h" -//#include "arm_correlate_fast_q15.h" - //using namespace std; /* DEBUG FUNCTION @@ -35,6 +33,8 @@ //----------VARIABLES HERE const int dataLength = 1000; const int captureLength = 50; +int dL = 1000; +int cL = 50; double temp = 22; double hum = 10; double micDist = 0.250; //meters @@ -192,10 +192,635 @@ } else return false; } +void luke_correlate_f32( + float* pSrcA, + int srcALen, + float* pSrcB, + int srcBLen, + float* pDst) +{ + + +#ifndef ARM_MATH_CM0_FAMILY + + /* Run the below code for Cortex-M4 and Cortex-M3 */ + + float *pIn1; /* inputA pointer */ + float *pIn2; /* inputB pointer */ + float *pOut = pDst; /* output pointer */ + float *px; /* Intermediate inputA pointer */ + float *py; /* Intermediate inputB pointer */ + float *pSrc1; /* Intermediate pointers */ + float sum, acc0, acc1, acc2, acc3; /* Accumulators */ + float x0, x1, x2, x3, c0; /* temporary variables for holding input and coefficient values */ + int j, k = 0u, count, blkCnt, outBlockSize, blockSize1, blockSize2, blockSize3; /* loop counters */ + int32_t inc = 1; /* Destination address modifier */ + + + /* The algorithm implementation is based on the lengths of the inputs. */ + /* srcB is always made to slide across srcA. */ + /* So srcBLen is always considered as shorter or equal to srcALen */ + /* But CORR(x, y) is reverse of CORR(y, x) */ + /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */ + /* and the destination pointer modifier, inc is set to -1 */ + /* If srcALen > srcBLen, zero pad has to be done to srcB to make the two inputs of same length */ + /* But to improve the performance, + * we assume zeroes in the output instead of zero padding either of the the inputs*/ + /* If srcALen > srcBLen, + * (srcALen - srcBLen) zeroes has to included in the starting of the output buffer */ + /* If srcALen < srcBLen, + * (srcALen - srcBLen) zeroes has to included in the ending of the output buffer */ + if(srcALen >= srcBLen) + { + /* Initialization of inputA pointer */ + pIn1 = pSrcA; + + /* Initialization of inputB pointer */ + pIn2 = pSrcB; + + /* Number of output samples is calculated */ + outBlockSize = (2u * srcALen) - 1u; + + /* When srcALen > srcBLen, zero padding has to be done to srcB + * to make their lengths equal. + * Instead, (outBlockSize - (srcALen + srcBLen - 1)) + * number of output samples are made zero */ + j = outBlockSize - (srcALen + (srcBLen - 1u)); + + /* Updating the pointer position to non zero value */ + pOut += j; + + //while(j > 0u) + //{ + // /* Zero is stored in the destination buffer */ + // *pOut++ = 0.0f; + + // /* Decrement the loop counter */ + // j--; + //} + + } + else + { + /* Initialization of inputA pointer */ + pIn1 = pSrcB; + + /* Initialization of inputB pointer */ + pIn2 = pSrcA; + + /* srcBLen is always considered as shorter or equal to srcALen */ + j = srcBLen; + srcBLen = srcALen; + srcALen = j; + + /* CORR(x, y) = Reverse order(CORR(y, x)) */ + /* Hence set the destination pointer to point to the last output sample */ + pOut = pDst + ((srcALen + srcBLen) - 2u); + + /* Destination address modifier is set to -1 */ + inc = -1; + + } + + /* The function is internally + * divided into three parts according to the number of multiplications that has to be + * taken place between inputA samples and inputB samples. In the first part of the + * algorithm, the multiplications increase by one for every iteration. + * In the second part of the algorithm, srcBLen number of multiplications are done. + * In the third part of the algorithm, the multiplications decrease by one + * for every iteration.*/ + /* The algorithm is implemented in three stages. + * The loop counters of each stage is initiated here. */ + blockSize1 = srcBLen - 1u; + blockSize2 = srcALen - (srcBLen - 1u); + blockSize3 = blockSize1; + + /* -------------------------- + * Initializations of stage1 + * -------------------------*/ + + /* sum = x[0] * y[srcBlen - 1] + * sum = x[0] * y[srcBlen-2] + x[1] * y[srcBlen - 1] + * .... + * sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen - 1] * y[srcBLen - 1] + */ + + /* In this stage the MAC operations are increased by 1 for every iteration. + The count variable holds the number of MAC operations performed */ + count = 1u; + + /* Working pointer of inputA */ + px = pIn1; + + /* Working pointer of inputB */ + pSrc1 = pIn2 + (srcBLen - 1u); + py = pSrc1; + + /* ------------------------ + * Stage1 process + * ----------------------*/ + + /* The first stage starts here */ + while(blockSize1 > 0u) + { + /* Accumulator is made zero for every iteration */ + sum = 0.0f; + + /* Apply loop unrolling and compute 4 MACs simultaneously. */ + 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] * y[srcBLen - 4] */ + sum += *px++ * *py++; + /* x[1] * y[srcBLen - 3] */ + sum += *px++ * *py++; + /* x[2] * y[srcBLen - 2] */ + sum += *px++ * *py++; + /* x[3] * y[srcBLen - 1] */ + sum += *px++ * *py++; + + /* Decrement the loop counter */ + k--; + } + + /* If the count is not a multiple of 4, compute any remaining MACs here. + ** No loop unrolling is used. */ + k = count % 0x4u; + + while(k > 0u) + { + /* Perform the multiply-accumulate */ + /* x[0] * y[srcBLen - 1] */ + sum += *px++ * *py++; + + /* Decrement the loop counter */ + k--; + } + + /* Store the result in the accumulator in the destination buffer. */ + *pOut = sum; + /* Destination pointer is updated according to the address modifier, inc */ + pOut += inc; + + /* Update the inputA and inputB pointers for next MAC calculation */ + py = pSrc1 - count; + px = pIn1; + + /* Increment the MAC count */ + count++; + + /* Decrement the loop counter */ + blockSize1--; + } + + /* -------------------------- + * Initializations of stage2 + * ------------------------*/ + + /* sum = x[0] * y[0] + x[1] * y[1] +...+ x[srcBLen-1] * y[srcBLen-1] + * sum = x[1] * y[0] + x[2] * y[1] +...+ x[srcBLen] * y[srcBLen-1] + * .... + * sum = x[srcALen-srcBLen-2] * y[0] + x[srcALen-srcBLen-1] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] + */ + + /* Working pointer of inputA */ + px = pIn1; + + /* Working pointer of inputB */ + py = pIn2; + + /* count is index by which the pointer pIn1 to be incremented */ + count = 0u; + + /* ------------------- + * Stage2 process + * ------------------*/ + + /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed. + * So, to loop unroll over blockSize2, + * srcBLen should be greater than or equal to 4, to loop unroll the srcBLen loop */ + if(srcBLen >= 4u) + { + /* Loop unroll over blockSize2, by 4 */ + blkCnt = blockSize2 >> 2u; + + while(blkCnt > 0u) + { + /* Set all accumulators to zero */ + acc0 = 0.0f; + acc1 = 0.0f; + acc2 = 0.0f; + acc3 = 0.0f; + + /* read x[0], x[1], x[2] samples */ + x0 = *(px++); + x1 = *(px++); + x2 = *(px++); + + /* Apply loop unrolling and compute 4 MACs simultaneously. */ + 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. */ + do + { + /* Read y[0] sample */ + c0 = *(py++); + + /* Read x[3] sample */ + x3 = *(px++); + + /* Perform the multiply-accumulate */ + /* acc0 += x[0] * y[0] */ + acc0 += x0 * c0; + /* acc1 += x[1] * y[0] */ + acc1 += x1 * c0; + /* acc2 += x[2] * y[0] */ + acc2 += x2 * c0; + /* acc3 += x[3] * y[0] */ + acc3 += x3 * c0; + + /* Read y[1] sample */ + c0 = *(py++); + + /* Read x[4] sample */ + x0 = *(px++); + + /* Perform the multiply-accumulate */ + /* acc0 += x[1] * y[1] */ + acc0 += x1 * c0; + /* acc1 += x[2] * y[1] */ + acc1 += x2 * c0; + /* acc2 += x[3] * y[1] */ + acc2 += x3 * c0; + /* acc3 += x[4] * y[1] */ + acc3 += x0 * c0; + + /* Read y[2] sample */ + c0 = *(py++); + + /* Read x[5] sample */ + x1 = *(px++); + + /* Perform the multiply-accumulates */ + /* acc0 += x[2] * y[2] */ + acc0 += x2 * c0; + /* acc1 += x[3] * y[2] */ + acc1 += x3 * c0; + /* acc2 += x[4] * y[2] */ + acc2 += x0 * c0; + /* acc3 += x[5] * y[2] */ + acc3 += x1 * c0; + + /* Read y[3] sample */ + c0 = *(py++); + + /* Read x[6] sample */ + x2 = *(px++); + + /* Perform the multiply-accumulates */ + /* acc0 += x[3] * y[3] */ + acc0 += x3 * c0; + /* acc1 += x[4] * y[3] */ + acc1 += x0 * c0; + /* acc2 += x[5] * y[3] */ + acc2 += x1 * c0; + /* acc3 += x[6] * y[3] */ + acc3 += x2 * c0; + + + } while(--k); + + /* If the srcBLen is not a multiple of 4, compute any remaining MACs here. + ** No loop unrolling is used. */ + k = srcBLen % 0x4u; + + while(k > 0u) + { + /* Read y[4] sample */ + c0 = *(py++); + + /* Read x[7] sample */ + x3 = *(px++); + + /* Perform the multiply-accumulates */ + /* acc0 += x[4] * y[4] */ + acc0 += x0 * c0; + /* acc1 += x[5] * y[4] */ + acc1 += x1 * c0; + /* acc2 += x[6] * y[4] */ + acc2 += x2 * c0; + /* acc3 += x[7] * y[4] */ + acc3 += x3 * c0; + + /* Reuse the present samples for the next MAC */ + x0 = x1; + x1 = x2; + x2 = x3; + + /* Decrement the loop counter */ + k--; + } + + /* Store the result in the accumulator in the destination buffer. */ + *pOut = acc0; + /* Destination pointer is updated according to the address modifier, inc */ + pOut += inc; + + *pOut = acc1; + pOut += inc; + + *pOut = acc2; + pOut += inc; + + *pOut = acc3; + pOut += inc; + + /* Increment the pointer pIn1 index, count by 4 */ + count += 4u; + + /* Update the inputA and inputB pointers for next MAC calculation */ + px = pIn1 + count; + py = pIn2; + + /* Decrement the loop counter */ + blkCnt--; + } + + /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here. + ** No loop unrolling is used. */ + blkCnt = blockSize2 % 0x4u; + + while(blkCnt > 0u) + { + /* Accumulator is made zero for every iteration */ + sum = 0.0f; + + /* Apply loop unrolling and compute 4 MACs simultaneously. */ + 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) + { + /* Perform the multiply-accumulates */ + sum += *px++ * *py++; + sum += *px++ * *py++; + sum += *px++ * *py++; + sum += *px++ * *py++; + + /* Decrement the loop counter */ + k--; + } + + /* If the srcBLen is not a multiple of 4, compute any remaining MACs here. + ** No loop unrolling is used. */ + k = srcBLen % 0x4u; + + while(k > 0u) + { + /* Perform the multiply-accumulate */ + sum += *px++ * *py++; + + /* Decrement the loop counter */ + k--; + } + + /* Store the result in the accumulator in the destination buffer. */ + *pOut = sum; + /* Destination pointer is updated according to the address modifier, inc */ + pOut += inc; + + /* Increment the pointer pIn1 index, count by 1 */ + count++; + + /* Update the inputA and inputB pointers for next MAC calculation */ + px = pIn1 + count; + py = pIn2; + + /* Decrement the loop counter */ + blkCnt--; + } + } + else + { + /* If the srcBLen is not a multiple of 4, + * the blockSize2 loop cannot be unrolled by 4 */ + blkCnt = blockSize2; + + while(blkCnt > 0u) + { + /* Accumulator is made zero for every iteration */ + sum = 0.0f; + + /* Loop over srcBLen */ + k = srcBLen; + + while(k > 0u) + { + /* Perform the multiply-accumulate */ + sum += *px++ * *py++; + + /* Decrement the loop counter */ + k--; + } + + /* Store the result in the accumulator in the destination buffer. */ + *pOut = sum; + /* Destination pointer is updated according to the address modifier, inc */ + pOut += inc; + + /* Increment the pointer pIn1 index, count by 1 */ + count++; + + /* Update the inputA and inputB pointers for next MAC calculation */ + px = pIn1 + count; + py = pIn2; + + /* Decrement the loop counter */ + blkCnt--; + } + } + + /* -------------------------- + * Initializations of stage3 + * -------------------------*/ + + /* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] + * sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1] + * .... + * sum += x[srcALen-2] * y[0] + x[srcALen-1] * y[1] + * sum += x[srcALen-1] * y[0] + */ + + /* In this stage the MAC operations are decreased by 1 for every iteration. + The count variable holds the number of MAC operations performed */ + count = srcBLen - 1u; + + /* Working pointer of inputA */ + pSrc1 = pIn1 + (srcALen - (srcBLen - 1u)); + px = pSrc1; + + /* Working pointer of inputB */ + py = pIn2; + + /* ------------------- + * Stage3 process + * ------------------*/ + + while(blockSize3 > 0u) + { + /* Accumulator is made zero for every iteration */ + sum = 0.0f; + + /* Apply loop unrolling and compute 4 MACs simultaneously. */ + 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) + { + /* Perform the multiply-accumulates */ + /* sum += x[srcALen - srcBLen + 4] * y[3] */ + sum += *px++ * *py++; + /* sum += x[srcALen - srcBLen + 3] * y[2] */ + sum += *px++ * *py++; + /* sum += x[srcALen - srcBLen + 2] * y[1] */ + sum += *px++ * *py++; + /* sum += x[srcALen - srcBLen + 1] * y[0] */ + sum += *px++ * *py++; + + /* Decrement the loop counter */ + k--; + } + + /* If the count is not a multiple of 4, compute any remaining MACs here. + ** No loop unrolling is used. */ + k = count % 0x4u; + + while(k > 0u) + { + /* Perform the multiply-accumulates */ + sum += *px++ * *py++; + + /* Decrement the loop counter */ + k--; + } + + /* Store the result in the accumulator in the destination buffer. */ + *pOut = sum; + /* Destination pointer is updated according to the address modifier, inc */ + pOut += inc; + + /* Update the inputA and inputB pointers for next MAC calculation */ + px = ++pSrc1; + py = pIn2; + + /* Decrement the MAC count */ + count--; + + /* Decrement the loop counter */ + blockSize3--; + } + +#else + + /* Run the below code for Cortex-M0 */ + + float *pIn1 = pSrcA; /* inputA pointer */ + float *pIn2 = pSrcB + (srcBLen - 1u); /* inputB pointer */ + float sum; /* Accumulator */ + int i = 0u, j; /* loop counters */ + int inv = 0u; /* Reverse order flag */ + int tot = 0u; /* Length */ + + /* The algorithm implementation is based on the lengths of the inputs. */ + /* srcB is always made to slide across srcA. */ + /* So srcBLen is always considered as shorter or equal to srcALen */ + /* But CORR(x, y) is reverse of CORR(y, x) */ + /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */ + /* and a varaible, inv is set to 1 */ + /* If lengths are not equal then zero pad has to be done to make the two + * inputs of same length. But to improve the performance, we assume zeroes + * in the output instead of zero padding either of the the inputs*/ + /* If srcALen > srcBLen, (srcALen - srcBLen) zeroes has to included in the + * starting of the output buffer */ + /* If srcALen < srcBLen, (srcALen - srcBLen) zeroes has to included in the + * ending of the output buffer */ + /* Once the zero padding is done the remaining of the output is calcualted + * using convolution but with the shorter signal time shifted. */ + + /* Calculate the length of the remaining sequence */ + tot = ((srcALen + srcBLen) - 2u); + + if(srcALen > srcBLen) + { + /* Calculating the number of zeros to be padded to the output */ + j = srcALen - srcBLen; + + /* Initialise the pointer after zero padding */ + pDst += j; + } + + else if(srcALen < srcBLen) + { + /* Initialization to inputB pointer */ + pIn1 = pSrcB; + + /* Initialization to the end of inputA pointer */ + pIn2 = pSrcA + (srcALen - 1u); + + /* Initialisation of the pointer after zero padding */ + pDst = pDst + tot; + + /* Swapping the lengths */ + j = srcALen; + srcALen = srcBLen; + srcBLen = j; + + /* Setting the reverse flag */ + inv = 1; + + } + + /* Loop to calculate convolution for output length number of times */ + for (i = 0u; i <= tot; i++) + { + /* Initialize sum with zero to carry on MAC operations */ + sum = 0.0f; + + /* Loop to perform MAC operations according to convolution equation */ + for (j = 0u; j <= i; j++) + { + /* Check the array limitations */ + if((((i - j) < srcBLen) && (j < srcALen))) + { + /* z[i] += x[i-j] * y[j] */ + sum += pIn1[j] * pIn2[-((int32_t) i - j)]; + } + } + /* Store the output in the destination buffer */ + if(inv == 1) + *pDst-- = sum; + else + *pDst++ = sum; + } + +#endif /* #ifndef ARM_MATH_CM0_FAMILY */ + +} + +/** + * @} end of Corr group + */ // main() runs in its own thread in the OS int main() { + for(int i = 0; i < test; i++) { delaytest[i] = -420 + i*105; } @@ -297,8 +922,12 @@ case CALC: DebugPrintState( std::cout << "Nucleo state is CALC: " << std::endl; ); //Debug( wait(0.5); ); - luke_correlate_f32(p1, captureLength, p2, captureLength, p_res); - + luke_correlate_f32(p1, cL, p2, cL, p_res); + Debug( + for(int i = 0; i < dataLength; i++){ + std::cout << dsp_res[i] << " "; + }); + int positionOfMaxVal_1 = FindPeak(1); int positionOfMaxVal_2 = FindPeak(2); //run functions
diff -r 3f71950ceb71 -r a89e836d9faf luke_correlate_f32.c --- a/luke_correlate_f32.c Fri May 12 06:59:55 2017 +0000 +++ b/luke_correlate_f32.c Fri May 12 08:01:26 2017 +0000 @@ -115,11 +115,11 @@ */ void luke_correlate_f32( - float * pSrcA, + float* pSrcA, int srcALen, - float * pSrcB, + float* pSrcB, int srcBLen, - float * pDst) + float* pDst) {
diff -r 3f71950ceb71 -r a89e836d9faf luke_correlate_f32.h --- a/luke_correlate_f32.h Fri May 12 06:59:55 2017 +0000 +++ b/luke_correlate_f32.h Fri May 12 08:01:26 2017 +0000 @@ -1,7 +1,7 @@ -#ifndef arm_correlate_f32_h // To make sure you don't declare the function more than once by including the header multiple times. -#define arm_correlate_f32_h +#ifndef luke_correlate_f32_h // To make sure you don't declare the function more than once by including the header multiple times. +#define luke_correlate_f32_h -void luke_correlate_f32(float * pSrcA, int srcALen, float * pSrcB, int srcBLen, float * pDst); +void luke_correlate_f32(float* pSrcA, int srcALen, float* pSrcB, int srcBLen, float* pDst); #endif \ No newline at end of file