RMUL2025/lib/cmsis_5/CMSIS/DSP/Testing/Source/Tests/FIRF16.cpp

165 lines
3.9 KiB
C++
Executable File

#include "FIRF16.h"
#include <stdio.h>
#include "Error.h"
#define SNR_THRESHOLD 60
/*
Reference patterns are generated with
a double precision computation.
*/
#define REL_ERROR (2.0e-2)
#if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
static __ALIGNED(8) float16_t coeffArray[32];
#endif
static void checkInnerTail(float16_t *b)
{
ASSERT_TRUE(b[0] == 0.0f);
ASSERT_TRUE(b[1] == 0.0f);
ASSERT_TRUE(b[2] == 0.0f);
ASSERT_TRUE(b[3] == 0.0f);
}
// Coef must be padded to a multiple of 4
#define FIRCOEFPADDING 2
void FIRF16::test_fir_f16()
{
const int16_t *configp = configs.ptr();
float16_t *statep = state.ptr();
const float16_t *orgcoefsp = coefs.ptr();
const float16_t *coefsp;
const float16_t *inputp = inputs.ptr();
float16_t *outp = output.ptr();
unsigned long i;
#if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
int j;
#endif
int blockSize;
int numTaps;
#if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
int round;
#endif
/*
Python script is generating different tests with
different blockSize and numTaps.
We loop on those configs.
*/
for(i=0; i < configs.nbSamples() ; i += 2)
{
blockSize = configp[0];
numTaps = configp[1];
#if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
/* Copy coefficients and pad to zero
*/
memset(coeffArray,127,32*sizeof(float16_t));
round = numTaps >> FIRCOEFPADDING;
if ((round << FIRCOEFPADDING) < numTaps)
{
round ++;
}
round = round<<FIRCOEFPADDING;
memset(coeffArray,0,round*sizeof(float16_t));
//printf("blockSize=%d, numTaps=%d, round=%d (%d)\n",blockSize,numTaps,round,round - numTaps);
for(j=0;j < numTaps; j++)
{
coeffArray[j] = orgcoefsp[j];
}
coefsp = coeffArray;
#else
coefsp = orgcoefsp;
#endif
/*
The filter is initialized with the coefs, blockSize and numTaps.
*/
arm_fir_init_f16(&this->S,numTaps,coefsp,statep,blockSize);
/*
Input pointer is reset since the same input pattern is used
*/
inputp = inputs.ptr();
/*
Python script is filtering a 2*blockSize number of samples.
We do the same filtering in two pass to check (indirectly that
the state management of the fir is working.)
*/
arm_fir_f16(&this->S,inputp,outp,blockSize);
outp += blockSize;
checkInnerTail(outp);
inputp += blockSize;
arm_fir_f16(&this->S,inputp,outp,blockSize);
outp += blockSize;
checkInnerTail(outp);
configp += 2;
orgcoefsp += numTaps;
}
ASSERT_EMPTY_TAIL(output);
ASSERT_SNR(output,ref,(float16_t)SNR_THRESHOLD);
ASSERT_REL_ERROR(output,ref,REL_ERROR);
}
void FIRF16::setUp(Testing::testID_t id,std::vector<Testing::param_t>& params,Client::PatternMgr *mgr)
{
(void)params;
switch(id)
{
case FIRF16::TEST_FIR_F16_1:
break;
}
inputs.reload(FIRF16::FIRINPUTS_F16_ID,mgr);
coefs.reload(FIRF16::FIRCOEFS_F16_ID,mgr);
configs.reload(FIRF16::FIRCONFIGS_S16_ID,mgr);
ref.reload(FIRF16::FIRREFS_F16_ID,mgr);
output.create(ref.nbSamples(),FIRF16::OUT_F16_ID,mgr);
/* > Max 8*ceil(blockSize,8) + blockSize + numTaps - 1 as generated by Python script */
state.create(47+47,FIRF16::OUT_F16_ID,mgr);
}
void FIRF16::tearDown(Testing::testID_t id,Client::PatternMgr *mgr)
{
(void)id;
output.dump(mgr);
}