4 |
This is a benchmark for comparison between a integer math, table lookup |
This is a benchmark for comparison between a integer math, table lookup |
5 |
and numeric sine wave harmonics solution. |
and numeric sine wave harmonics solution. |
6 |
|
|
7 |
Copyright (C) 2005 Christian Schoenebeck <cuse@users.sf.net> |
Copyright (C) 2005 - 2017 Christian Schoenebeck <cuse@users.sf.net> |
8 |
*/ |
*/ |
9 |
|
|
10 |
#include <math.h> |
#include <math.h> |
11 |
#include <time.h> |
#include <time.h> |
12 |
#include <stdio.h> |
#include <stdio.h> |
13 |
|
|
14 |
|
#include "../src/engines/common/LFOTriangleIntMath.h" |
15 |
|
#include "../src/engines/common/LFOTriangleIntAbsMath.h" |
16 |
|
#include "../src/engines/common/LFOTriangleDiHarmonic.h" |
17 |
|
|
18 |
// whether we should not show any messages on the console |
// whether we should not show any messages on the console |
19 |
#ifndef SILENT |
#ifndef SILENT |
20 |
# define SILENT 0 |
# define SILENT 0 |
42 |
# define SIGNED 1 |
# define SIGNED 1 |
43 |
#endif |
#endif |
44 |
|
|
45 |
|
// maximum value of the LFO output (also depends on SIGNED above) |
46 |
|
#ifndef MAX |
47 |
|
# define MAX 1.0f |
48 |
|
#endif |
49 |
|
|
50 |
// pro forma |
// pro forma |
51 |
#ifndef SAMPLING_RATE |
#ifndef SAMPLING_RATE |
52 |
# define SAMPLING_RATE 44100.0f |
# define SAMPLING_RATE 44100.0f |
57 |
#define INT_MATH_SOLUTION 2 /* we don't start with 1, as this is reserved for unknown errors */ |
#define INT_MATH_SOLUTION 2 /* we don't start with 1, as this is reserved for unknown errors */ |
58 |
#define DI_HARMONIC_SOLUTION 3 |
#define DI_HARMONIC_SOLUTION 3 |
59 |
#define TABLE_LOOKUP_SOLUTION 4 /* table lookup solution is currently disabled in this benchmark, see below */ |
#define TABLE_LOOKUP_SOLUTION 4 /* table lookup solution is currently disabled in this benchmark, see below */ |
60 |
|
#define INT_MATH_ABS_SOLUTION 5 /* integer math with abs() */ |
61 |
#define INVALID_RESULT -1 |
#define INVALID_RESULT -1 |
62 |
|
|
63 |
// we use 32 bit single precision floating point as sample point format |
// we use 32 bit single precision floating point as sample point format |
64 |
typedef float sample_t; |
typedef float smpl_t; // (sample_t is already defined as int16_t by global_private.h) |
65 |
|
|
66 |
|
using namespace LinuxSampler; |
67 |
|
|
68 |
|
#if SIGNED |
69 |
|
LFOTriangleIntMath<range_signed>* pIntLFO = NULL; |
70 |
|
LFOTriangleIntAbsMath<range_signed>* pIntAbsLFO = NULL; |
71 |
|
LFOTriangleDiHarmonic<range_signed>* pDiHarmonicLFO = NULL; |
72 |
|
#else // unsigned |
73 |
|
LFOTriangleIntMath<range_unsigned>* pIntLFO = NULL; |
74 |
|
LFOTriangleIntAbsMath<range_unsigned>* pIntAbsLFO = NULL; |
75 |
|
LFOTriangleDiHarmonic<range_unsigned>* pDiHarmonicLFO = NULL; |
76 |
|
#endif |
77 |
|
|
78 |
// integer math solution |
// integer math solution |
79 |
float int_math(sample_t* pDestinationBuffer, float* pAmp, const int steps, const float frequency) { |
float int_math(smpl_t* pDestinationBuffer, float* pAmp, const int steps, const float frequency) { |
80 |
// pro forma |
// pro forma |
81 |
const float r = frequency / SAMPLING_RATE; // frequency alteration quotient |
pIntLFO->trigger(frequency, start_level_max, 1200 /* max. internal depth */, 0, false, (unsigned int) SAMPLING_RATE); |
82 |
unsigned int maxvalue = (unsigned int) -1; // all 0xFFFF... |
|
83 |
#if SIGNED |
clock_t stop_time; |
84 |
const float normalizer = 1.0f / ((float) maxvalue / 4.0f); |
clock_t start_time = clock(); |
85 |
#else // unsigned |
|
86 |
const float normalizer = 1.0f / ((float) maxvalue / 2.0f); |
for (int run = 0; run < RUNS; run++) { |
87 |
|
pIntLFO->updateByMIDICtrlValue(127); // pro forma |
88 |
|
for (int i = 0; i < steps; ++i) { |
89 |
|
pDestinationBuffer[i] = pIntLFO->render() * pAmp[i]; // * pAmp[i] just to simulate some memory load |
90 |
|
} |
91 |
|
} |
92 |
|
|
93 |
|
stop_time = clock(); |
94 |
|
float elapsed_time = (stop_time - start_time) / (double(CLOCKS_PER_SEC) / 1000.0); |
95 |
|
#if ! SILENT |
96 |
|
printf("int math solution elapsed time: %1.0f ms\n", elapsed_time); |
97 |
#endif |
#endif |
|
const int c = (int) (maxvalue * r); |
|
|
const int signshifts = (sizeof(int) * 8) - 1; |
|
98 |
|
|
99 |
int iSign; |
return elapsed_time; |
100 |
|
} |
101 |
|
|
102 |
|
// integer math abs solution |
103 |
|
float int_math_abs(smpl_t* pDestinationBuffer, float* pAmp, const int steps, const float frequency) { |
104 |
|
// pro forma |
105 |
|
pIntAbsLFO->trigger(frequency, start_level_max, 1200 /* max. internal depth */, 0, false, (unsigned int) SAMPLING_RATE); |
106 |
|
|
107 |
clock_t stop_time; |
clock_t stop_time; |
108 |
clock_t start_time = clock(); |
clock_t start_time = clock(); |
109 |
|
|
110 |
for (int run = 0; run < RUNS; run++) { |
for (int run = 0; run < RUNS; run++) { |
111 |
int iLevel = 0; |
pIntAbsLFO->updateByMIDICtrlValue(0); // pro forma |
112 |
for (int i = 0; i < steps; ++i) { |
for (int i = 0; i < steps; ++i) { |
113 |
iLevel += c; |
pDestinationBuffer[i] = pIntAbsLFO->render() * pAmp[i]; // * pAmp[i] just to simulate some memory load |
|
iSign = (iLevel >> signshifts) | 1; |
|
|
#if SIGNED |
|
|
pDestinationBuffer[i] = (normalizer * (float) (iSign * iLevel) - 1.0f) * pAmp[i]; // * pAmp[i] just to simulate some memory load |
|
|
#else // unsigned |
|
|
pDestinationBuffer[i] = normalizer * (float) (iSign * iLevel) * pAmp[i]; // * pAmp[i] just to simulate some memory load |
|
|
#endif |
|
114 |
} |
} |
115 |
} |
} |
116 |
|
|
117 |
stop_time = clock(); |
stop_time = clock(); |
118 |
float elapsed_time = (stop_time - start_time) / (double(CLOCKS_PER_SEC) / 1000.0); |
float elapsed_time = (stop_time - start_time) / (double(CLOCKS_PER_SEC) / 1000.0); |
119 |
#if ! SILENT |
#if ! SILENT |
120 |
printf("int math solution elapsed time: %1.0f ms\n", elapsed_time); |
printf("int math abs solution elapsed time: %1.0f ms\n", elapsed_time); |
121 |
#endif |
#endif |
122 |
|
|
123 |
return elapsed_time; |
return elapsed_time; |
131 |
// anyway. If you found an architecture where this seems to be the best |
// anyway. If you found an architecture where this seems to be the best |
132 |
// solution, please let us know! |
// solution, please let us know! |
133 |
#if 0 |
#if 0 |
134 |
float table_lookup(sample_t* pDestinationBuffer, float* pAmp, const int steps, const float frequency) { |
float table_lookup(smpl_t* pDestinationBuffer, float* pAmp, const int steps, const float frequency) { |
135 |
// pro forma |
// pro forma |
136 |
const float r = frequency / SAMPLING_RATE; // frequency alteration quotient |
const float r = frequency / SAMPLING_RATE; // frequency alteration quotient |
137 |
#if SIGNED |
#if SIGNED |
194 |
#endif |
#endif |
195 |
|
|
196 |
// numeric, di-harmonic solution |
// numeric, di-harmonic solution |
197 |
float numeric_di_harmonic_solution(sample_t* pDestinationBuffer, float* pAmp, const int steps, const float frequency) { |
float numeric_di_harmonic_solution(smpl_t* pDestinationBuffer, float* pAmp, const int steps, const float frequency) { |
198 |
// we approximate the triangular wave by adding 2 harmonics |
// pro forma |
199 |
const float c1 = 2.0f * M_PI * frequency / SAMPLING_RATE; |
pDiHarmonicLFO->trigger(frequency, start_level_max, 1200 /* max. internal depth */, 0, false, (unsigned int) SAMPLING_RATE); |
|
const float phi1 = 0.0f; // phase displacement |
|
|
const float c2 = 2.0f * M_PI * frequency / SAMPLING_RATE * 3.0f; |
|
|
const float phi2 = 0.0f; // phase displacement |
|
|
|
|
|
// amplitue for the 2nd harmonic (to approximate the triangular wave) |
|
|
const float amp2 = 0.1f; |
|
|
|
|
|
// initial values for real and imaginary part |
|
|
float real1 = cos(phi1); |
|
|
float imag1 = sin(phi1); |
|
|
float real2 = cos(phi2); |
|
|
float imag2 = sin(phi2); |
|
200 |
|
|
201 |
clock_t stop_time; |
clock_t stop_time; |
202 |
clock_t start_time = clock(); |
clock_t start_time = clock(); |
203 |
|
|
204 |
for (int run = 0; run < RUNS; run++) { |
for (int run = 0; run < RUNS; run++) { |
205 |
for (int i = 0; i < steps; i++) { |
pDiHarmonicLFO->updateByMIDICtrlValue(127); // pro forma |
206 |
#if SIGNED |
for (int i = 0; i < steps; ++i) { |
207 |
pDestinationBuffer[i] = (real1 + real2 * amp2) * pAmp[i]; // * pAmp[i] just to simulate some memory load |
pDestinationBuffer[i] = pDiHarmonicLFO->render() * pAmp[i]; // * pAmp[i] just to simulate some memory load |
|
#else // unsigned |
|
|
pDestinationBuffer[i] = ((real1 + real2 * amp2) * 0.5f + 0.5f) * pAmp[i]; // * pAmp[i] just to simulate some memory load |
|
|
#endif |
|
|
real1 -= c1 * imag1; |
|
|
imag1 += c1 * real1; |
|
|
real2 -= c2 * imag2; |
|
|
imag2 += c2 * real2; |
|
208 |
} |
} |
209 |
} |
} |
210 |
|
|
218 |
} |
} |
219 |
|
|
220 |
// output calculated values as RAW audio format (32 bit floating point, mono) file |
// output calculated values as RAW audio format (32 bit floating point, mono) file |
221 |
void output_as_raw_file(const char* filename, sample_t* pOutputBuffer, int steps) { |
void output_as_raw_file(const char* filename, smpl_t* pOutputBuffer, int steps) { |
222 |
FILE* file = fopen(filename, "w"); |
FILE* file = fopen(filename, "w"); |
223 |
if (file) { |
if (file) { |
224 |
fwrite((void*) pOutputBuffer, sizeof(float), steps, file); |
fwrite((void*) pOutputBuffer, sizeof(float), steps, file); |
242 |
# endif |
# endif |
243 |
#endif |
#endif |
244 |
|
|
245 |
|
#if SIGNED |
246 |
|
pIntLFO = new LFOTriangleIntMath<range_signed>(MAX); |
247 |
|
pIntAbsLFO = new LFOTriangleIntAbsMath<range_signed>(MAX); |
248 |
|
pDiHarmonicLFO = new LFOTriangleDiHarmonic<range_signed>(MAX); |
249 |
|
#else // unsigned |
250 |
|
pIntLFO = new LFOTriangleIntMath<range_unsigned>(MAX); |
251 |
|
pIntAbsLFO = new LFOTriangleIntAbsMath<range_unsigned>(MAX); |
252 |
|
pDiHarmonicLFO = new LFOTriangleDiHarmonic<range_unsigned>(MAX); |
253 |
|
#endif |
254 |
|
|
255 |
// output buffer for the calculated sinusoid wave |
// output buffer for the calculated sinusoid wave |
256 |
sample_t* pOutputBuffer = new sample_t[steps]; |
smpl_t* pOutputBuffer = new smpl_t[steps]; |
257 |
// just an arbitrary amplitude envelope to simulate a bit higher memory bandwidth |
// just an arbitrary amplitude envelope to simulate a bit higher memory bandwidth |
258 |
float* pAmplitude = new float[steps]; |
float* pAmplitude = new float[steps]; |
259 |
|
|
262 |
pAmplitude[i] = (float) i / (float) steps; |
pAmplitude[i] = (float) i / (float) steps; |
263 |
|
|
264 |
// how long each solution took (in seconds) |
// how long each solution took (in seconds) |
265 |
float int_math_result, /*table_lookup_result,*/ numeric_di_harmonic_result; |
float int_math_result, int_math_abs_result, /*table_lookup_result,*/ numeric_di_harmonic_result; |
266 |
|
|
267 |
int_math_result = int_math(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
int_math_result = int_math(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
268 |
#if OUTPUT_AS_RAW_WAVE |
#if OUTPUT_AS_RAW_WAVE |
269 |
output_as_raw_file("bench_int_math.raw", pOutputBuffer, steps); |
output_as_raw_file("bench_int_math.raw", pOutputBuffer, steps); |
270 |
#endif |
#endif |
271 |
|
|
272 |
|
int_math_abs_result = int_math_abs(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
273 |
|
#if OUTPUT_AS_RAW_WAVE |
274 |
|
output_as_raw_file("bench_int_math_abs.raw", pOutputBuffer, steps); |
275 |
|
#endif |
276 |
//table_lookup_result = table_lookup(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
//table_lookup_result = table_lookup(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
277 |
//#if OUTPUT_AS_RAW_WAVE |
//#if OUTPUT_AS_RAW_WAVE |
278 |
// output_as_raw_file("bench_table_lookup.raw", pOutputBuffer, steps); |
// output_as_raw_file("bench_table_lookup.raw", pOutputBuffer, steps); |
287 |
#endif |
#endif |
288 |
|
|
289 |
int_math_result += int_math(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
int_math_result += int_math(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
290 |
|
int_math_abs_result = int_math_abs(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
291 |
//table_lookup_result += table_lookup(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
//table_lookup_result += table_lookup(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
292 |
numeric_di_harmonic_result += numeric_di_harmonic_solution(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
numeric_di_harmonic_result += numeric_di_harmonic_solution(pOutputBuffer, pAmplitude, steps, sinusoidFrequency); |
293 |
|
|
294 |
if (pAmplitude) delete[] pAmplitude; |
if (pAmplitude) delete[] pAmplitude; |
295 |
if (pOutputBuffer) delete[] pOutputBuffer; |
if (pOutputBuffer) delete[] pOutputBuffer; |
296 |
|
|
297 |
if (/*int_math_result <= table_lookup_result &&*/ int_math_result <= numeric_di_harmonic_result) return INT_MATH_SOLUTION; |
if (pIntLFO) delete pIntLFO; |
298 |
if (/*numeric_di_harmonic_result <= table_lookup_result &&*/ numeric_di_harmonic_result <= int_math_result) return DI_HARMONIC_SOLUTION; |
if (pDiHarmonicLFO) delete pDiHarmonicLFO; |
299 |
//if (table_lookup_result <= int_math_result && table_lookup_result <= numeric_di_harmonic_result) return TABLE_LOOKUP_SOLUTION; |
|
300 |
|
if (int_math_abs_result <= int_math_result && int_math_abs_result <= numeric_di_harmonic_result) return INT_MATH_ABS_SOLUTION; |
301 |
|
if (int_math_result <= int_math_abs_result && int_math_result <= numeric_di_harmonic_result) return INT_MATH_SOLUTION; |
302 |
|
if (numeric_di_harmonic_result <= int_math_abs_result && numeric_di_harmonic_result <= int_math_result) return DI_HARMONIC_SOLUTION; |
303 |
|
|
304 |
return INVALID_RESULT; // error |
return INVALID_RESULT; // error |
305 |
} |
} |