dengzhihao
/
CMSIS-DSP
0
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

CMSIS-DSP: Added benchmarking code for Transform.

Browse Source
pull/19/head
Christophe Favergeon 6 years ago
parent 5a6f1e2baa
commit 799b7aa521
20 changed files with 16361 additions and 15 deletions
Show Stats Download Patch File Download Diff File

2
Source/TransformFunctions/CMakeLists.txt
Unescape Escape View File

@ -110,7 +110,7 @@ target_sources(CMSISDSPTransform PRIVATE arm_cfft_q31.c)
target_sources(CMSISDSPTransform PRIVATE arm_cfft_radix4_q31.c)
endif()
# For scipy or or wrappers
# For scipy or wrappers or benchmarks
if (WRAPPER)
target_sources(CMSISDSPTransform PRIVATE arm_cfft_radix2_init_f32.c)
endif()

3
Testing/CMakeLists.txt
Unescape Escape View File

@ -118,6 +118,9 @@ set(TESTSRC
Source/Benchmarks/BinaryF32.cpp
Source/Benchmarks/BinaryQ31.cpp
Source/Benchmarks/BinaryQ15.cpp
Source/Benchmarks/TransformF32.cpp
Source/Benchmarks/TransformQ31.cpp
Source/Benchmarks/TransformQ15.cpp
Source/Benchmarks/FullyConnectedBench.cpp
Source/Benchmarks/PoolingBench.cpp
)

32
Testing/Include/Benchmarks/TransformF32.h
Unescape Escape View File

@ -0,0 +1,32 @@
#include "Test.h"
#include "Pattern.h"
class TransformF32:public Client::Suite
{
public:
TransformF32(Testing::testID_t id);
void setUp(Testing::testID_t,std::vector<Testing::param_t>& params,Client::PatternMgr *mgr);
void tearDown(Testing::testID_t,Client::PatternMgr *mgr);
private:
#include "TransformF32_decl.h"
Client::Pattern<float32_t> samples;
Client::LocalPattern<float32_t> output;
Client::LocalPattern<float32_t> state;
int nbSamples;
int ifft;
int bitRev;
float32_t *pSrc;
float32_t *pDst;
float32_t *pState;
const arm_cfft_instance_f32 *cfftInstance;
arm_rfft_fast_instance_f32 rfftFastInstance;
arm_dct4_instance_f32 dct4Instance;
arm_rfft_instance_f32 rfftInstance;
arm_cfft_radix4_instance_f32 cfftRadix4Instance;
arm_cfft_radix2_instance_f32 cfftRadix2Instance;
};

31
Testing/Include/Benchmarks/TransformQ15.h
Unescape Escape View File

@ -0,0 +1,31 @@
#include "Test.h"
#include "Pattern.h"
class TransformQ15:public Client::Suite
{
public:
TransformQ15(Testing::testID_t id);
void setUp(Testing::testID_t,std::vector<Testing::param_t>& params,Client::PatternMgr *mgr);
void tearDown(Testing::testID_t,Client::PatternMgr *mgr);
private:
#include "TransformQ15_decl.h"
Client::Pattern<q15_t> samples;
Client::LocalPattern<q15_t> output;
Client::LocalPattern<q15_t> state;
int nbSamples;
int ifft;
int bitRev;
q15_t *pSrc;
q15_t *pDst;
q15_t *pState;
arm_cfft_instance_q15 cfftInstance;
arm_dct4_instance_q15 dct4Instance;
arm_rfft_instance_q15 rfftInstance;
arm_cfft_radix4_instance_q15 cfftRadix4Instance;
arm_cfft_radix2_instance_q15 cfftRadix2Instance;
};

31
Testing/Include/Benchmarks/TransformQ31.h
Unescape Escape View File

@ -0,0 +1,31 @@
#include "Test.h"
#include "Pattern.h"
class TransformQ31:public Client::Suite
{
public:
TransformQ31(Testing::testID_t id);
void setUp(Testing::testID_t,std::vector<Testing::param_t>& params,Client::PatternMgr *mgr);
void tearDown(Testing::testID_t,Client::PatternMgr *mgr);
private:
#include "TransformQ31_decl.h"
Client::Pattern<q31_t> samples;
Client::LocalPattern<q31_t> output;
Client::LocalPattern<q31_t> state;
int nbSamples;
int ifft;
int bitRev;
q31_t *pSrc;
q31_t *pDst;
q31_t *pState;
arm_cfft_instance_q31 cfftInstance;
arm_dct4_instance_q31 dct4Instance;
arm_rfft_instance_q31 rfftInstance;
arm_cfft_radix4_instance_q31 cfftRadix4Instance;
arm_cfft_radix2_instance_q31 cfftRadix2Instance;
};

51
Testing/PatternGeneration/Transform.py
Unescape Escape View File

@ -0,0 +1,51 @@
import os.path
import numpy as np
import itertools
import Tools
# Those patterns are used for tests and benchmarks.
# For tests, there is the need to add tests for saturation
def randComplex(nb):
data = np.random.randn(2*nb)
data = data/max(data)
data_comp = data.view(dtype=np.complex128)
return(data_comp)
def asReal(a):
#return(a.view(dtype=np.float64))
return(a.reshape(np.size(a)).view(dtype=np.float64))
def writeTests(config):
NBRSAMPLES=2048
NBCSAMPLES=256
samples=np.random.randn(NBRSAMPLES)
samples = np.abs(samples/max(samples))
samplesC=randComplex(NBCSAMPLES)
config.writeInput(1, samples,"RealSamples")
config.writeInput(1, asReal(samplesC),"ComplexSamples")
PATTERNDIR = os.path.join("Patterns","DSP","Transform","Transform")
PARAMDIR = os.path.join("Parameters","DSP","Transform","Transform")
configf32=Tools.Config(PATTERNDIR,PARAMDIR,"f32")
configq31=Tools.Config(PATTERNDIR,PARAMDIR,"q31")
configq15=Tools.Config(PATTERNDIR,PARAMDIR,"q15")
writeTests(configf32)
writeTests(configq31)
writeTests(configq15)

1026
Testing/Patterns/DSP/Transform/TransformF32/ComplexSamples1_f32.txt
View File

File diff suppressed because it is too large Load Diff

4098
Testing/Patterns/DSP/Transform/TransformF32/RealSamples1_f32.txt
View File

File diff suppressed because it is too large Load Diff

1026
Testing/Patterns/DSP/Transform/TransformQ15/ComplexSamples1_q15.txt
View File

File diff suppressed because it is too large Load Diff

4098
Testing/Patterns/DSP/Transform/TransformQ15/RealSamples1_q15.txt
View File

File diff suppressed because it is too large Load Diff

1026
Testing/Patterns/DSP/Transform/TransformQ31/ComplexSamples1_q31.txt
View File

File diff suppressed because it is too large Load Diff

4098
Testing/Patterns/DSP/Transform/TransformQ31/RealSamples1_q31.txt
View File

File diff suppressed because it is too large Load Diff

166
Testing/Source/Benchmarks/TransformF32.cpp
Unescape Escape View File

@ -0,0 +1,166 @@
#include "TransformF32.h"
#include "Error.h"
#include "arm_const_structs.h"
const arm_cfft_instance_f32 *arm_cfft_get_instance_f32(uint16_t fftLen)
{
switch (fftLen) {
case 16:
return(&arm_cfft_sR_f32_len16);
break;
case 32:
return(&arm_cfft_sR_f32_len32);
break;
case 64:
return(&arm_cfft_sR_f32_len64);
break;
case 128:
return(&arm_cfft_sR_f32_len128);
break;
case 256:
return(&arm_cfft_sR_f32_len256);
break;
case 512:
return(&arm_cfft_sR_f32_len512);
break;
case 1024:
return(&arm_cfft_sR_f32_len1024);
break;
case 2048:
return(&arm_cfft_sR_f32_len2048);
break;
case 4096:
return(&arm_cfft_sR_f32_len4096);
break;
}
return(NULL);
}
void TransformF32::test_cfft_f32()
{
arm_cfft_f32(this->cfftInstance, this->pDst, this->ifft,this->bitRev);
}
void TransformF32::test_rfft_f32()
{
arm_rfft_fast_f32(&this->rfftFastInstance, this->pSrc, this->pDst, this->ifft);
}
void TransformF32::test_dct4_f32()
{
arm_dct4_f32(
&this->dct4Instance,
this->pState,
this->pDst);
}
void TransformF32::test_cfft_radix4_f32()
{
arm_cfft_radix4_f32(&this->cfftRadix4Instance,this->pDst);
}
void TransformF32::test_cfft_radix2_f32()
{
arm_cfft_radix2_f32(&this->cfftRadix2Instance,this->pDst);
}
void TransformF32::setUp(Testing::testID_t id,std::vector<Testing::param_t>& params,Client::PatternMgr *mgr)
{
float32_t normalize;
std::vector<Testing::param_t>::iterator it = params.begin();
this->nbSamples = *it++;
this->ifft = *it++;
this->bitRev = *it;
switch(id)
{
case TEST_CFFT_F32_1:
samples.reload(TransformF32::INPUTC_F32_ID,mgr,2*this->nbSamples);
output.create(2*this->nbSamples,TransformF32::OUT_F32_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
this->cfftInstance=arm_cfft_get_instance_f32(this->nbSamples);
memcpy(this->pDst,this->pSrc,2*sizeof(float32_t)*this->nbSamples);
break;
case TEST_RFFT_F32_2:
samples.reload(TransformF32::INPUTR_F32_ID,mgr,this->nbSamples);
output.create(this->nbSamples,TransformF32::OUT_F32_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
arm_rfft_fast_init_f32(&this->rfftFastInstance, this->nbSamples);
break;
case TEST_DCT4_F32_3:
samples.reload(TransformF32::INPUTR_F32_ID,mgr,this->nbSamples);
output.create(this->nbSamples,TransformF32::OUT_F32_ID,mgr);
state.create(2*this->nbSamples,TransformF32::STATE_F32_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
this->pState=state.ptr();
normalize = sqrt((2.0f/(float32_t)this->nbSamples));
memcpy(this->pDst,this->pSrc,sizeof(float32_t)*this->nbSamples);
arm_dct4_init_f32(
&this->dct4Instance,
&this->rfftInstance,
&this->cfftRadix4Instance,
this->nbSamples,
this->nbSamples/2,
normalize);
break;
case TEST_CFFT_RADIX4_F32_4:
samples.reload(TransformF32::INPUTC_F32_ID,mgr,2*this->nbSamples);
output.create(2*this->nbSamples,TransformF32::OUT_F32_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
memcpy(this->pDst,this->pSrc,2*sizeof(float32_t)*this->nbSamples);
arm_cfft_radix4_init_f32(&this->cfftRadix4Instance,
this->nbSamples,
this->ifft,
this->bitRev);
break;
case TEST_CFFT_RADIX2_F32_5:
samples.reload(TransformF32::INPUTC_F32_ID,mgr,2*this->nbSamples);
output.create(2*this->nbSamples,TransformF32::OUT_F32_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
memcpy(this->pDst,this->pSrc,2*sizeof(float32_t)*this->nbSamples);
arm_cfft_radix2_init_f32(&this->cfftRadix2Instance,
this->nbSamples,
this->ifft,
this->bitRev);
break;
}
}
void TransformF32::tearDown(Testing::testID_t id,Client::PatternMgr *mgr)
{
}

224
Testing/Source/Benchmarks/TransformQ15.cpp
Unescape Escape View File

@ -0,0 +1,224 @@
#include "TransformQ15.h"
#include "Error.h"
#include "arm_const_structs.h"
#define CFFT_INIT(typ, suffix, S, fftLen) \
\
{ \
\
/* Initialise the default arm status */ \
arm_status status = ARM_MATH_SUCCESS; \
\
/* Initialise the FFT length */ \
S->fftLen = fftLen; \
\
/* Initialise the Twiddle coefficient pointer */ \
S->pTwiddle = (typ *)twiddleCoef_4096_##suffix; \
\
\
/* Initializations of Instance structure depending on the FFT length */\
switch (S->fftLen) { \
\
\
/* Initializations of structure parameters for 2048 point FFT */ \
case 2048U: \
/* Initialise the bit reversal table modifier */ \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_2048_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_2048; \
S->pTwiddle = (typ *)twiddleCoef_2048_##suffix; \
break; \
\
/* Initializations of structure parameters for 1024 point FFT */ \
case 1024U: \
/* Initialise the bit reversal table modifier */ \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_1024_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_1024; \
S->pTwiddle = (typ *)twiddleCoef_1024_##suffix; \
break; \
\
/* Initializations of structure parameters for 512 point FFT */ \
case 512U: \
/* Initialise the bit reversal table modifier */ \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_512_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_512; \
S->pTwiddle = (typ *)twiddleCoef_512_##suffix; \
break; \
\
case 256U: \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_256_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_256; \
S->pTwiddle = (typ *)twiddleCoef_256_##suffix; \
\
break; \
\
case 128U: \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_128_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_128; \
S->pTwiddle = (typ *)twiddleCoef_128_##suffix; \
\
break; \
\
case 64U: \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_64_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_64; \
S->pTwiddle = (typ *)twiddleCoef_64_##suffix; \
break; \
\
case 32U: \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_32_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_32; \
S->pTwiddle = (typ *)twiddleCoef_32_##suffix; \
break; \
\
case 16U: \
/* Initializations of structure parameters for 16 point FFT */ \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_16_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_16; \
S->pTwiddle = (typ *)twiddleCoef_16_##suffix; \
break; \
\
\
default: \
/* Reporting argument error if fftSize is not valid value */ \
status = ARM_MATH_ARGUMENT_ERROR; \
break; \
} \
\
\
return (status); \
\
}
arm_status arm_cfft_init_q15(arm_cfft_instance_q15 *S, uint16_t fftLen)
{
CFFT_INIT(q15_t, q15, S, fftLen);
}
void TransformQ15::test_cfft_q15()
{
arm_cfft_q15(&this->cfftInstance, this->pDst, this->ifft,this->bitRev);
}
void TransformQ15::test_rfft_q15()
{
arm_rfft_q15(&this->rfftInstance, this->pSrc, this->pDst);
}
void TransformQ15::test_dct4_q15()
{
arm_dct4_q15(
&this->dct4Instance,
this->pState,
this->pDst);
}
void TransformQ15::test_cfft_radix4_q15()
{
arm_cfft_radix4_q15(&this->cfftRadix4Instance,this->pDst);
}
void TransformQ15::test_cfft_radix2_q15()
{
arm_cfft_radix2_q15(&this->cfftRadix2Instance,this->pDst);
}
void TransformQ15::setUp(Testing::testID_t id,std::vector<Testing::param_t>& params,Client::PatternMgr *mgr)
{
float32_t normalize;
std::vector<Testing::param_t>::iterator it = params.begin();
this->nbSamples = *it++;
this->ifft = *it++;
this->bitRev = *it;
switch(id)
{
case TEST_CFFT_Q15_1:
samples.reload(TransformQ15::INPUTC_Q15_ID,mgr,2*this->nbSamples);
output.create(2*this->nbSamples,TransformQ15::OUT_Q15_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
arm_cfft_init_q15(&this->cfftInstance,this->nbSamples);
memcpy(this->pDst,this->pSrc,2*sizeof(q15_t)*this->nbSamples);
break;
case TEST_RFFT_Q15_2:
samples.reload(TransformQ15::INPUTR_Q15_ID,mgr,this->nbSamples);
output.create(this->nbSamples,TransformQ15::OUT_Q15_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
arm_rfft_init_q15(&this->rfftInstance, this->nbSamples, this->ifft, this->bitRev);
break;
case TEST_DCT4_Q15_3:
samples.reload(TransformQ15::INPUTR_Q15_ID,mgr,this->nbSamples);
output.create(this->nbSamples,TransformQ15::OUT_Q15_ID,mgr);
state.create(2*this->nbSamples,TransformQ15::STATE_Q15_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
this->pState=state.ptr();
normalize = sqrt((2.0f/(float32_t)this->nbSamples));
memcpy(this->pDst,this->pSrc,sizeof(q15_t)*this->nbSamples);
arm_dct4_init_q15(
&this->dct4Instance,
&this->rfftInstance,
&this->cfftRadix4Instance,
this->nbSamples,
this->nbSamples/2,
normalize);
break;
case TEST_CFFT_RADIX4_Q15_4:
samples.reload(TransformQ15::INPUTC_Q15_ID,mgr,2*this->nbSamples);
output.create(2*this->nbSamples,TransformQ15::OUT_Q15_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
memcpy(this->pDst,this->pSrc,2*sizeof(q15_t)*this->nbSamples);
arm_cfft_radix4_init_q15(&this->cfftRadix4Instance,
this->nbSamples,
this->ifft,
this->bitRev);
break;
case TEST_CFFT_RADIX2_Q15_5:
samples.reload(TransformQ15::INPUTC_Q15_ID,mgr,2*this->nbSamples);
output.create(2*this->nbSamples,TransformQ15::OUT_Q15_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
memcpy(this->pDst,this->pSrc,2*sizeof(q15_t)*this->nbSamples);
arm_cfft_radix2_init_q15(&this->cfftRadix2Instance,
this->nbSamples,
this->ifft,
this->bitRev);
break;
}
}
void TransformQ15::tearDown(Testing::testID_t id,Client::PatternMgr *mgr)
{
}

223
Testing/Source/Benchmarks/TransformQ31.cpp
Unescape Escape View File

@ -0,0 +1,223 @@
#include "TransformQ31.h"
#include "Error.h"
#include "arm_const_structs.h"
#define CFFT_INIT(typ, suffix, S, fftLen) \
\
{ \
\
/* Initialise the default arm status */ \
arm_status status = ARM_MATH_SUCCESS; \
\
/* Initialise the FFT length */ \
S->fftLen = fftLen; \
\
/* Initialise the Twiddle coefficient pointer */ \
S->pTwiddle = (typ *)twiddleCoef_4096_##suffix; \
\
\
/* Initializations of Instance structure depending on the FFT length */\
switch (S->fftLen) { \
\
/* Initializations of structure parameters for 2048 point FFT */ \
case 2048U: \
/* Initialise the bit reversal table modifier */ \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_2048_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_2048; \
S->pTwiddle = (typ *)twiddleCoef_2048_##suffix; \
break; \
\
/* Initializations of structure parameters for 1024 point FFT */ \
case 1024U: \
/* Initialise the bit reversal table modifier */ \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_1024_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_1024; \
S->pTwiddle = (typ *)twiddleCoef_1024_##suffix; \
break; \
\
/* Initializations of structure parameters for 512 point FFT */ \
case 512U: \
/* Initialise the bit reversal table modifier */ \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_512_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_512; \
S->pTwiddle = (typ *)twiddleCoef_512_##suffix; \
break; \
\
case 256U: \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_256_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_256; \
S->pTwiddle = (typ *)twiddleCoef_256_##suffix; \
\
break; \
\
case 128U: \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_128_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_128; \
S->pTwiddle = (typ *)twiddleCoef_128_##suffix; \
\
break; \
\
case 64U: \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_64_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_64; \
S->pTwiddle = (typ *)twiddleCoef_64_##suffix; \
break; \
\
case 32U: \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_32_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_32; \
S->pTwiddle = (typ *)twiddleCoef_32_##suffix; \
break; \
\
case 16U: \
/* Initializations of structure parameters for 16 point FFT */ \
S->bitRevLength = ARMBITREVINDEXTABLE_FIXED_16_TABLE_LENGTH; \
S->pBitRevTable = (uint16_t *)armBitRevIndexTable_fixed_16; \
S->pTwiddle = (typ *)twiddleCoef_16_##suffix; \
break; \
\
\
default: \
/* Reporting argument error if fftSize is not valid value */ \
status = ARM_MATH_ARGUMENT_ERROR; \
break; \
} \
\
\
return (status); \
\
}
arm_status arm_cfft_init_q31(arm_cfft_instance_q31 *S, uint16_t fftLen)
{
CFFT_INIT(q31_t, q31, S, fftLen);
}
void TransformQ31::test_cfft_q31()
{
arm_cfft_q31(&this->cfftInstance, this->pDst, this->ifft,this->bitRev);
}
void TransformQ31::test_rfft_q31()
{
arm_rfft_q31(&this->rfftInstance, this->pSrc, this->pDst);
}
void TransformQ31::test_dct4_q31()
{
arm_dct4_q31(
&this->dct4Instance,
this->pState,
this->pDst);
}
void TransformQ31::test_cfft_radix4_q31()
{
arm_cfft_radix4_q31(&this->cfftRadix4Instance,this->pDst);
}
void TransformQ31::test_cfft_radix2_q31()
{
arm_cfft_radix2_q31(&this->cfftRadix2Instance,this->pDst);
}
void TransformQ31::setUp(Testing::testID_t id,std::vector<Testing::param_t>& params,Client::PatternMgr *mgr)
{
float32_t normalize;
std::vector<Testing::param_t>::iterator it = params.begin();
this->nbSamples = *it++;
this->ifft = *it++;
this->bitRev = *it;
switch(id)
{
case TEST_CFFT_Q31_1:
samples.reload(TransformQ31::INPUTC_Q31_ID,mgr,2*this->nbSamples);
output.create(2*this->nbSamples,TransformQ31::OUT_Q31_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
arm_cfft_init_q31(&this->cfftInstance,this->nbSamples);
memcpy(this->pDst,this->pSrc,2*sizeof(q31_t)*this->nbSamples);
break;
case TEST_RFFT_Q31_2:
samples.reload(TransformQ31::INPUTR_Q31_ID,mgr,this->nbSamples);
output.create(this->nbSamples,TransformQ31::OUT_Q31_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
arm_rfft_init_q31(&this->rfftInstance, this->nbSamples,this->ifft,this->bitRev);
break;
case TEST_DCT4_Q31_3:
samples.reload(TransformQ31::INPUTR_Q31_ID,mgr,this->nbSamples);
output.create(this->nbSamples,TransformQ31::OUT_Q31_ID,mgr);
state.create(2*this->nbSamples,TransformQ31::STATE_Q31_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
this->pState=state.ptr();
normalize = sqrt((2.0f/(float32_t)this->nbSamples));
memcpy(this->pDst,this->pSrc,sizeof(q31_t)*this->nbSamples);
arm_dct4_init_q31(
&this->dct4Instance,
&this->rfftInstance,
&this->cfftRadix4Instance,
this->nbSamples,
this->nbSamples/2,
normalize);
break;
case TEST_CFFT_RADIX4_Q31_4:
samples.reload(TransformQ31::INPUTC_Q31_ID,mgr,2*this->nbSamples);
output.create(2*this->nbSamples,TransformQ31::OUT_Q31_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
memcpy(this->pDst,this->pSrc,2*sizeof(q31_t)*this->nbSamples);
arm_cfft_radix4_init_q31(&this->cfftRadix4Instance,
this->nbSamples,
this->ifft,
this->bitRev);
break;
case TEST_CFFT_RADIX2_Q31_5:
samples.reload(TransformQ31::INPUTC_Q31_ID,mgr,2*this->nbSamples);
output.create(2*this->nbSamples,TransformQ31::OUT_Q31_ID,mgr);
this->pSrc=samples.ptr();
this->pDst=output.ptr();
memcpy(this->pDst,this->pSrc,2*sizeof(q31_t)*this->nbSamples);
arm_cfft_radix2_init_q31(&this->cfftRadix2Instance,
this->nbSamples,
this->ifft,
this->bitRev);
break;
}
}
void TransformQ31::tearDown(Testing::testID_t id,Client::PatternMgr *mgr)
{
}

2
Testing/addAllBenchToDatabase.bat
Unescape Escape View File

@ -22,3 +22,5 @@ echo "Unary Matrix"
python addToDB.py -f bench.txt Unary
echo "Binary Matrix"
python addToDB.py -f bench.txt Binary
echo "Transform"
python addToDB.py -f bench.txt Transform

2
Testing/addAllBenchToRegressionDatabase.bat
Unescape Escape View File

@ -22,3 +22,5 @@ echo "Unary Matrix"
python addToRegDB.py -f bench.txt Unary
echo "Binary Matrix"
python addToRegDB.py -f bench.txt Binary
echo "Transform"
python addToRegDB.py -f bench.txt Transform

137
Testing/bench.txt
Unescape Escape View File

@ -1270,6 +1270,143 @@ group Root {
}
}
group Transform {
class = Transform
folder = Transform
suite Transform F32 {
class = TransformF32
folder = TransformF32
ParamList {
NB,IFFT,BITREV
Summary NB
Names "Number of samples,Inverse FFT, Bit Reversal"
Formula "NB"
}
Pattern INPUTR_F32_ID : RealSamples1_f32.txt
Pattern INPUTC_F32_ID : ComplexSamples1_f32.txt
Output OUT_F32_ID : Output
Output STATE_F32_ID : Output
Params CFFT_PARAM_ID = {
NB = [16,64,128,256]
IFFT = [0,1]
REV = [0,1]
}
Params RFFT_PARAM_ID = {
NB = [32,64,128,256]
IFFT = [0,1]
REV = [1]
}
Params DCT_PARAM_ID = {
NB = [128,512,2048]
IFFT = [0]
REV = [1]
}
Functions {
test_cfft_f32:test_cfft_f32 -> CFFT_PARAM_ID
test_rfft_f32:test_rfft_f32 -> RFFT_PARAM_ID
test_dct4_f32:test_dct4_f32 -> DCT_PARAM_ID
test_cfft_radix4_f32:test_cfft_radix4_f32 -> CFFT_PARAM_ID
test_cfft_radix2_f32:test_cfft_radix2_f32 -> CFFT_PARAM_ID
}
}
suite Transform Q31 {
class = TransformQ31
folder = TransformQ31
ParamList {
NB,IFFT,BITREV
Summary NB
Names "Number of samples,Inverse FFT, Bit Reversal"
Formula "NB"
}
Pattern INPUTR_Q31_ID : RealSamples1_q31.txt
Pattern INPUTC_Q31_ID : ComplexSamples1_q31.txt
Output OUT_Q31_ID : Output
Output STATE_Q31_ID : Output
Params CFFT_PARAM_ID = {
NB = [16,64,128,256]
IFFT = [0,1]
REV = [0,1]
}
Params RFFT_PARAM_ID = {
NB = [32,64,128,256]
IFFT = [0,1]
REV = [0,1]
}
Params DCT_PARAM_ID = {
NB = [128,512,2048]
IFFT = [0]
REV = [1]
}
Functions {
test_cfft_q31:test_cfft_q31 -> CFFT_PARAM_ID
test_rfft_q31:test_rfft_q31 -> RFFT_PARAM_ID
test_dct4_q31:test_dct4_q31 -> DCT_PARAM_ID
test_cfft_radix4_q31:test_cfft_radix4_q31 -> CFFT_PARAM_ID
test_cfft_radix2_q31:test_cfft_radix2_q31 -> CFFT_PARAM_ID
}
}
suite Transform Q15 {
class = TransformQ15
folder = TransformQ15
ParamList {
NB,IFFT,BITREV
Summary NB
Names "Number of samples,Inverse FFT, Bit Reversal"
Formula "NB"
}
Pattern INPUTR_Q15_ID : RealSamples1_q15.txt
Pattern INPUTC_Q15_ID : ComplexSamples1_q15.txt
Output OUT_Q15_ID : Output
Output STATE_Q15_ID : Output
Params CFFT_PARAM_ID = {
NB = [16,64,128,256]
IFFT = [0,1]
REV = [0,1]
}
Params RFFT_PARAM_ID = {
NB = [32,64,128,256]
IFFT = [0,1]
REV = [1]
}
Params DCT_PARAM_ID = {
NB = [128,512,2048]
IFFT = [0]
REV = [1]
}
Functions {
test_cfft_q15:test_cfft_q15 -> CFFT_PARAM_ID
test_rfft_q15:test_rfft_q15 -> RFFT_PARAM_ID
test_dct4_q15:test_dct4_q15 -> DCT_PARAM_ID
test_cfft_radix4_q15:test_cfft_radix4_q15 -> CFFT_PARAM_ID
test_cfft_radix2_q15:test_cfft_radix2_q15 -> CFFT_PARAM_ID
}
}
}
}
group NN Benchmarks

26
Testing/generateAllRegressions.bat
Unescape Escape View File

@ -1,24 +1,26 @@
echo "Basic Maths"
python summaryBench.py -f bench.txt BasicBenchmarks
python summaryBench.py -f bench.txt -r build_m7\result.txt BasicBenchmarks
echo "Complex Maths"
python summaryBench.py -f bench.txt ComplexBenchmarks
python summaryBench.py -f bench.txt -r build_m7\result.txt ComplexBenchmarks
echo "FIR"
python summaryBench.py -f bench.txt FIR
python summaryBench.py -f bench.txt -r build_m7\result.txt FIR
echo "Convolution / Correlation"
python summaryBench.py -f bench.txt MISC
python summaryBench.py -f bench.txt -r build_m7\result.txt MISC
echo "Decimation / Interpolation"
python summaryBench.py -f bench.txt DECIM
python summaryBench.py -f bench.txt -r build_m7\result.txt DECIM
echo "BiQuad"
python summaryBench.py -f bench.txt BIQUAD
python summaryBench.py -f bench.txt -r build_m7\result.txt BIQUAD
echo "Controller"
python summaryBench.py -f bench.txt Controller
python summaryBench.py -f bench.txt -r build_m7\result.txt Controller
echo "Fast Math"
python summaryBench.py -f bench.txt FastMath
python summaryBench.py -f bench.txt -r build_m7\result.txt FastMath
echo "Barycenter"
python summaryBench.py -f bench.txt SupportBarF32
python summaryBench.py -f bench.txt -r build_m7\result.txt SupportBarF32
echo "Support"
python summaryBench.py -f bench.txt Support
python summaryBench.py -f bench.txt -r build_m7\result.txt Support
echo "Unary Matrix"
python summaryBench.py -f bench.txt Unary
python summaryBench.py -f bench.txt -r build_m7\result.txt Unary
echo "Binary Matrix"
python summaryBench.py -f bench.txt Binary
python summaryBench.py -f bench.txt -r build_m7\result.txt Binary
echo "Transform"
python summaryBench.py -f bench.txt -r build_m7\result.txt Transform

70
Testing/runAllBenchmarks.bat
Unescape Escape View File

@ -0,0 +1,70 @@
@ECHO OFF
echo "Basic Maths"
python processTests.py -f bench.txt -e BasicBenchmarks
call:runBench
echo "Complex Maths"
python processTests.py -f bench.txt -e ComplexBenchmarks
call:runBench
echo "FIR"
python processTests.py -f bench.txt -e FIR
call:runBench
echo "Convolution / Correlation"
python processTests.py -f bench.txt -e MISC
call:runBench
echo "Decimation / Interpolation"
python processTests.py -f bench.txt -e DECIM
call:runBench
echo "BiQuad"
python processTests.py -f bench.txt -e BIQUAD
call:runBench
echo "Controller"
python processTests.py -f bench.txt -e Controller
call:runBench
echo "Fast Math"
python processTests.py -f bench.txt -e FastMath
call:runBench
echo "Barycenter"
python processTests.py -f bench.txt -e SupportBarF32
call:runBench
echo "Support"
python processTests.py -f bench.txt -e Support
call:runBench
echo "Unary Matrix"
python processTests.py -f bench.txt -e Unary
call:runBench
echo "Binary Matrix"
python processTests.py -f bench.txt -e Binary
call:runBench
echo "Transform"
python processTests.py -f bench.txt -e Transform
call:runBench
EXIT /B
:runBench
REM pushd build_m7
REM pushd build_m0
pushd build_a5
make
REM "C:\Program Files\ARM\Development Studio 2019.0\sw\models\bin\FVP_MPS2_Cortex-M7.exe" -a Testing > result.txt
REM "C:\Program Files\ARM\Development Studio 2019.0\sw\models\bin\FVP_MPS2_Cortex-M0.exe" -a Testing > result.txt
"C:\Program Files\ARM\Development Studio 2019.0\sw\models\bin\FVP_VE_Cortex-A5x1.exe" -a Testing > result.txt
popd
echo "Parse result"
REM python processResult.py -f bench.txt -e -r build_m7\result.txt
REM python processResult.py -f bench.txt -e -r build_m0\result.txt
python processResult.py -f bench.txt -e -r build_a5\result.txt
goto:eof
Write Preview
Loading…
Cancel
Save