mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2025-10-29 22:42:59 +00:00
Code Issues Releases Wiki Activity

Merge branch 'new_volk_module' of git+ssh://github.com/gnss-sdr/gnss-sdr

Browse Source 
into new_volk_module

# Conflicts:
#	src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_dot_prod_16ic.h
This commit is contained in:
Carles Fernandez
2016-01-19 00:01:26 +01:00
parent 1d18ff6c16 4ba75a3fbe
commit 090f6524db
5 changed files with 85 additions and 73 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_dot_prod_16ic.h
View File

@@ -40,7 +40,6 @@
#include <volk_gnsssdr/volk_gnsssdr_common.h> #include <volk_gnsssdr/volk_gnsssdr_common.h>
#include <volk_gnsssdr/volk_gnsssdr_complex.h> #include <volk_gnsssdr/volk_gnsssdr_complex.h>
#include <volk_gnsssdr/saturation_arithmetic.h> #include <volk_gnsssdr/saturation_arithmetic.h>
//#include <stdio.h>
#ifdef LV_HAVE_GENERIC #ifdef LV_HAVE_GENERIC
@@ -56,11 +55,9 @@ static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_generic(lv_16sc_t* result,
result[0] = lv_cmake((int16_t)0, (int16_t)0); result[0] = lv_cmake((int16_t)0, (int16_t)0);
for (unsigned int n = 0; n < num_points; n++) for (unsigned int n = 0; n < num_points; n++)
{ {
//r*a.r - i*a.i, i*a.r + r*a.i
lv_16sc_t tmp = in_a[n] * in_b[n]; lv_16sc_t tmp = in_a[n] * in_b[n];
result[0] = lv_cmake(sat_adds16i(lv_creal(result[0]), lv_creal(tmp)), sat_adds16i(lv_cimag(result[0]), lv_cimag(tmp) )); result[0] = lv_cmake(sat_adds16i(lv_creal(result[0]), lv_creal(tmp)), sat_adds16i(lv_cimag(result[0]), lv_cimag(tmp) ));
} }
//printf("generic result = %i,%i", lv_creal(result[0]),lv_cimag(result[0]));
} }
#endif /*LV_HAVE_GENERIC*/ #endif /*LV_HAVE_GENERIC*/
@@ -70,7 +67,6 @@ static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_generic(lv_16sc_t* result,
#include <emmintrin.h> #include <emmintrin.h>
static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_a_sse2(lv_16sc_t* out, const lv_16sc_t* in_a, const lv_16sc_t* in_b, unsigned int num_points) static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_a_sse2(lv_16sc_t* out, const lv_16sc_t* in_a, const lv_16sc_t* in_b, unsigned int num_points)
{ {
lv_16sc_t dotProduct = lv_cmake((int16_t)0, (int16_t)0); lv_16sc_t dotProduct = lv_cmake((int16_t)0, (int16_t)0);
const unsigned int sse_iters = num_points / 4; const unsigned int sse_iters = num_points / 4;

2
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_32fc_convert_16ic.h
View File

@@ -110,7 +110,7 @@ static inline void volk_gnsssdr_32fc_convert_16ic_u_sse(lv_16sc_t* outputVector,
float max_val = 32767; float max_val = 32767;
__m128 inputVal1, inputVal2; __m128 inputVal1, inputVal2;
__m128i intInputVal1, intInputVal2; __m128i intInputVal1, intInputVal2; // is __m128i defined in xmmintrin.h?
__m128 ret1, ret2; __m128 ret1, ret2;
__m128 vmin_val = _mm_set_ps1(min_val); __m128 vmin_val = _mm_set_ps1(min_val);
__m128 vmax_val = _mm_set_ps1(max_val); __m128 vmax_val = _mm_set_ps1(max_val);

11
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_8i_max_s8i.h
View File

@@ -77,7 +77,7 @@ static inline void volk_gnsssdr_8i_max_s8i_u_sse4_1(char* target, const char* sr
} }
} }
for(unsigned int i = 0; i<(num_points % 16); ++i) for(unsigned int i = sse_iters * 16; i< num_points; ++i)
{ {
if(src0[i] > max) if(src0[i] > max)
{ {
@@ -140,7 +140,7 @@ static inline void volk_gnsssdr_8i_max_s8i_u_sse2(char* target, const char* src0
inputPtr += 16; inputPtr += 16;
} }
for(unsigned int i = 0; i<(num_points % 16); ++i) for(unsigned int i = sse_iters * 16; i < num_points; ++i)
{ {
if(src0[i] > max) if(src0[i] > max)
{ {
@@ -228,7 +228,7 @@ static inline void volk_gnsssdr_8i_max_s8i_a_sse4_1(char* target, const char* sr
} }
} }
for(unsigned int i = 0; i<(num_points % 16); ++i) for(unsigned int i = sse_iters * 16; i < num_points; ++i)
{ {
if(src0[i] > max) if(src0[i] > max)
{ {
@@ -291,7 +291,7 @@ static inline void volk_gnsssdr_8i_max_s8i_a_sse2(char* target, const char* src0
inputPtr += 16; inputPtr += 16;
} }
for(unsigned int i = 0; i<(num_points % 16); ++i) for(unsigned int i = sse_iters * 16; i < num_points; ++i)
{ {
if(src0[i] > max) if(src0[i] > max)
{ {
@@ -313,8 +313,6 @@ static inline void volk_gnsssdr_8i_max_s8i_a_sse2(char* target, const char* src0
*/ */
static inline void volk_gnsssdr_8i_max_s8i_a_generic(char* target, const char* src0, unsigned int num_points) static inline void volk_gnsssdr_8i_max_s8i_a_generic(char* target, const char* src0, unsigned int num_points)
{ {
if(num_points > 0)
{
if(num_points > 0) if(num_points > 0)
{ {
char max = src0[0]; char max = src0[0];
@@ -327,7 +325,6 @@ static inline void volk_gnsssdr_8i_max_s8i_a_generic(char* target, const char* s
} }
target[0] = max; target[0] = max;
} }
}
} }
#endif /*LV_HAVE_GENERIC*/ #endif /*LV_HAVE_GENERIC*/

28
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/lib/qa_utils.cc
View File

@@ -39,7 +39,7 @@
#include <volk_gnsssdr/volk_gnsssdr_malloc.h> #include <volk_gnsssdr/volk_gnsssdr_malloc.h>
float uniform() { float uniform() {
return M_PI*2.0f * ((float) rand() / RAND_MAX - 0.5f); // uniformly (-1, 1) return 2.0f * ((float) rand() / RAND_MAX - 0.5f); // uniformly (-1, 1)
} }
template <class t> template <class t>
@@ -72,13 +72,11 @@ void load_random_data(void *data, volk_gnsssdr_type_t type, unsigned int n)
else ((uint64_t *)data)[i] = (uint64_t) scaled_rand; else ((uint64_t *)data)[i] = (uint64_t) scaled_rand;
break; break;
case 4: case 4:
if(type.is_signed) ((int32_t *)data)[i] = (int32_t) scaled_rand; if(type.is_signed) ((int32_t *)data)[i] = (int32_t) scaled_rand;
else ((uint32_t *)data)[i] = (uint32_t) scaled_rand; else ((uint32_t *)data)[i] = (uint32_t) scaled_rand;
break; break;
case 2: case 2:
// 16 bits dot product saturates very fast even with moderate lenght vectors // 16 bits dot product saturates very fast even with moderate length vectors
// we produce here only 4 bits input range // we produce here only 4 bits input range
if(type.is_signed) ((int16_t *)data)[i] = (int16_t)((int16_t) scaled_rand % 16); if(type.is_signed) ((int16_t *)data)[i] = (int16_t)((int16_t) scaled_rand % 16);
@@ -190,7 +188,7 @@ static void get_signatures_from_name(std::vector<volk_gnsssdr_type_t> &inputsig,
try { try {
multiplier = boost::lexical_cast<int>(token.substr(1, token.size()-1)); //will throw if invalid /////////// multiplier = boost::lexical_cast<int>(token.substr(1, token.size()-1)); //will throw if invalid ///////////
} catch(...) { } catch(...) {
multiplier = 1; // This is our '..._xn' mulriple correlator. Assign a fixed number here for the test? multiplier = 1; // This is our '..._xn' multiple correlator. Assign a fixed number here for the test?
} }
for(int i=1; i<multiplier; i++) { for(int i=1; i<multiplier; i++) {
if(side == SIDE_INPUT) inputsig.push_back(inputsig.back()); if(side == SIDE_INPUT) inputsig.push_back(inputsig.back());
@@ -221,6 +219,10 @@ inline void run_cast_test2(volk_gnsssdr_fn_2arg func, std::vector<void *> &buffs
while(iter--) func(buffs[0], buffs[1], vlen, arch.c_str()); while(iter--) func(buffs[0], buffs[1], vlen, arch.c_str());
} }
//inline void run_cast_test2(volk_gnsssdr_fn_2arg_r func, std::vector<void *> &buffs, float rem_code_phase_chips, float code_phase_step_chips, unsigned int vlen, int code_length_chips, unsigned int iter, std::string arch) {
// while(iter--) func(buffs[0], buffs[1], rem_code_phase_chips, code_phase_step_chips, vlen, code_length_chips, arch.c_str());
//}
inline void run_cast_test3(volk_gnsssdr_fn_3arg func, std::vector<void *> &buffs, unsigned int vlen, unsigned int iter, std::string arch) { inline void run_cast_test3(volk_gnsssdr_fn_3arg func, std::vector<void *> &buffs, unsigned int vlen, unsigned int iter, std::string arch) {
while(iter--) func(buffs[0], buffs[1], buffs[2], vlen, arch.c_str()); while(iter--) func(buffs[0], buffs[1], buffs[2], vlen, arch.c_str());
} }
@@ -291,16 +293,25 @@ inline void run_cast_test1_s16ic(volk_gnsssdr_fn_1arg_s16ic func, std::vector<vo
while(iter--) func(buffs[0], scalar, vlen, arch.c_str()); while(iter--) func(buffs[0], scalar, vlen, arch.c_str());
} }
inline void run_cast_test2_s16ic(volk_gnsssdr_fn_2arg_s16ic func, std::vector<void *> &buffs, lv_16sc_t scalar, unsigned int vlen, unsigned int iter, std::string arch) inline void run_cast_test2_s16ic(volk_gnsssdr_fn_2arg_s16ic func, std::vector<void *> &buffs, lv_16sc_t scalar, unsigned int vlen, unsigned int iter, std::string arch)
{ {
while(iter--) func(buffs[0], buffs[1], scalar, vlen, arch.c_str()); while(iter--) func(buffs[0], buffs[1], scalar, vlen, arch.c_str());
} }
//inline void run_cast_test2_s16ic(volk_gnsssdr_fn_2arg_s16ic func, std::vector<void *> &buffs, float rem_code_phase_chips, float code_phase_step_chips, int code_length_chips, unsigned int vlen, unsigned int iter, std::string arch)
//{
// while(iter--) func(buffs[0], buffs[1], rem_code_phase_chips, code_phase_step_chips, vlen, code_length_chips, arch.c_str());
//}
inline void run_cast_test3_s16ic(volk_gnsssdr_fn_3arg_s16ic func, std::vector<void *> &buffs, lv_16sc_t scalar, unsigned int vlen, unsigned int iter, std::string arch) inline void run_cast_test3_s16ic(volk_gnsssdr_fn_3arg_s16ic func, std::vector<void *> &buffs, lv_16sc_t scalar, unsigned int vlen, unsigned int iter, std::string arch)
{ {
while(iter--) func(buffs[0], buffs[1], buffs[2], scalar, vlen, arch.c_str()); while(iter--) func(buffs[0], buffs[1], buffs[2], scalar, vlen, arch.c_str());
} }
// end new // end new
inline void run_cast_test8(volk_gnsssdr_fn_8arg func, std::vector<void *> &buffs, unsigned int vlen, unsigned int iter, std::string arch) inline void run_cast_test8(volk_gnsssdr_fn_8arg func, std::vector<void *> &buffs, unsigned int vlen, unsigned int iter, std::string arch)
@@ -601,7 +612,14 @@ bool run_volk_gnsssdr_tests(volk_gnsssdr_func_desc_t desc,
case 2: case 2:
if(inputsc.size() == 0) if(inputsc.size() == 0)
{ {
// if(name.find("resampler") != std::string::npos)
// {
// run_cast_test2((volk_gnsssdr_fn_2arg_r)(manual_func), test_data[i], 1, 1, vlen, 1000, iter, arch_list[i]);
// }
// else
// {
run_cast_test2((volk_gnsssdr_fn_2arg)(manual_func), test_data[i], vlen, iter, arch_list[i]); run_cast_test2((volk_gnsssdr_fn_2arg)(manual_func), test_data[i], vlen, iter, arch_list[i]);
// }
} }
else if(inputsc.size() == 1 && inputsc[0].is_float) else if(inputsc.size() == 1 && inputsc[0].is_float)
{ {

1
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/lib/qa_utils.h
View File

@@ -151,6 +151,7 @@ bool run_volk_gnsssdr_tests(
#define VOLK_PUPPET_PROFILE(func, puppet_master_func, test_params, results) run_volk_gnsssdr_tests(func##_get_func_desc(), (void (*)())func##_manual, std::string(#func), test_params, results, std::string(#puppet_master_func)) #define VOLK_PUPPET_PROFILE(func, puppet_master_func, test_params, results) run_volk_gnsssdr_tests(func##_get_func_desc(), (void (*)())func##_manual, std::string(#func), test_params, results, std::string(#puppet_master_func))
typedef void (*volk_gnsssdr_fn_1arg)(void *, unsigned int, const char*); //one input, operate in place typedef void (*volk_gnsssdr_fn_1arg)(void *, unsigned int, const char*); //one input, operate in place
typedef void (*volk_gnsssdr_fn_2arg)(void *, void *, unsigned int, const char*); typedef void (*volk_gnsssdr_fn_2arg)(void *, void *, unsigned int, const char*);
//typedef void (*volk_gnsssdr_fn_2arg_r)(void *, void *, float, float, unsigned int, float, const char*);
typedef void (*volk_gnsssdr_fn_3arg)(void *, void *, void *, unsigned int, const char*); typedef void (*volk_gnsssdr_fn_3arg)(void *, void *, void *, unsigned int, const char*);
typedef void (*volk_gnsssdr_fn_4arg)(void *, void *, void *, void *, unsigned int, const char*); typedef void (*volk_gnsssdr_fn_4arg)(void *, void *, void *, void *, unsigned int, const char*);
typedef void (*volk_gnsssdr_fn_1arg_s32f)(void *, float, unsigned int, const char*); //one input vector, one scalar float input typedef void (*volk_gnsssdr_fn_1arg_s32f)(void *, float, unsigned int, const char*); //one input vector, one scalar float input