mirrors/pulseaudio
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/pulseaudio/pulseaudio.git synced 2025-11-07 13:30:03 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

tests: add cpu test

Browse source 
It tests only x86 architecture right now.
This commit is contained in:
Deng Zhengrong 2012-07-17 14:45:52 +08:00 • committed by Arun Raghavan
parent 6e33eb0073
commit c8cd89a7cb
7 changed files with 306 additions and 230 deletions
Download patch file Download diff file Expand all files Collapse all files

8
src/pulsecore/cpu-x86.c
View file

@ -44,7 +44,7 @@ static void get_cpuid(uint32_t op, uint32_t *a, uint32_t *b, uint32_t *c, uint32
}
#endif
pa_bool_t pa_cpu_init_x86(pa_cpu_x86_flag_t *flags) {
void pa_cpu_get_x86_flags(pa_cpu_x86_flag_t *flags) {
#if defined (__i386__) || defined (__amd64__)
uint32_t eax, ebx, ecx, edx;
uint32_t level;
@ -111,6 +111,12 @@ pa_bool_t pa_cpu_init_x86(pa_cpu_x86_flag_t *flags) {
(*flags & PA_CPU_X86_MMXEXT) ? "MMXEXT " : "",
(*flags & PA_CPU_X86_3DNOW) ? "3DNOW " : "",
(*flags & PA_CPU_X86_3DNOWEXT) ? "3DNOWEXT " : "");
#endif /* defined (__i386__) || defined (__amd64__) */
}
pa_bool_t pa_cpu_init_x86(pa_cpu_x86_flag_t *flags) {
#if defined (__i386__) || defined (__amd64__)
pa_cpu_get_x86_flags(flags);
/* activate various optimisations */
if (*flags & PA_CPU_X86_MMX) {

1
src/pulsecore/cpu-x86.h
View file

@ -40,6 +40,7 @@ typedef enum pa_cpu_x86_flag {
PA_CPU_X86_CMOV = (1 << 10)
} pa_cpu_x86_flag_t;
void pa_cpu_get_x86_flags(pa_cpu_x86_flag_t *flags);
pa_bool_t pa_cpu_init_x86 (pa_cpu_x86_flag_t *flags);
#if defined (__i386__)

54
src/pulsecore/sconv_sse.c
View file

@ -162,65 +162,11 @@ static void pa_sconv_s16le_from_f32ne_sse2(unsigned n, const float *a, int16_t *
);
}
#undef RUN_TEST
#ifdef RUN_TEST
#define SAMPLES 1019
#define TIMES 1000
static void run_test(void) {
int16_t samples[SAMPLES];
int16_t samples_ref[SAMPLES];
float floats[SAMPLES];
int i;
pa_usec_t start, stop;
pa_convert_func_t func;
printf("checking SSE %zd\n", sizeof(samples));
memset(samples_ref, 0, sizeof(samples_ref));
memset(samples, 0, sizeof(samples));
for (i = 0; i < SAMPLES; i++) {
floats[i] = 2.1f * (rand()/(float) RAND_MAX - 0.5f);
}
func = pa_get_convert_from_float32ne_function(PA_SAMPLE_S16LE);
func(SAMPLES, floats, samples_ref);
pa_sconv_s16le_from_f32ne_sse2(SAMPLES, floats, samples);
for (i = 0; i < SAMPLES; i++) {
if (samples[i] != samples_ref[i]) {
printf ("%d: %04x != %04x (%f)\n", i, samples[i], samples_ref[i],
floats[i]);
}
}
start = pa_rtclock_now();
for (i = 0; i < TIMES; i++) {
pa_sconv_s16le_from_f32ne_sse2(SAMPLES, floats, samples);
}
stop = pa_rtclock_now();
pa_log_info("SSE: %llu usec.", (long long unsigned int)(stop - start));
start = pa_rtclock_now();
for (i = 0; i < TIMES; i++) {
func(SAMPLES, floats, samples_ref);
}
stop = pa_rtclock_now();
pa_log_info("ref: %llu usec.", (long long unsigned int)(stop - start));
}
#endif
#endif /* defined (__i386__) || defined (__amd64__) */
void pa_convert_func_init_sse(pa_cpu_x86_flag_t flags) {
#if !defined(__APPLE__) && defined (__i386__) || defined (__amd64__)
#ifdef RUN_TEST
run_test();
#endif
if (flags & PA_CPU_X86_SSE2) {
pa_log_info("Initialising SSE2 optimized conversions.");
pa_set_convert_from_float32ne_function(PA_SAMPLE_S16LE, (pa_convert_func_t) pa_sconv_s16le_from_f32ne_sse2);

90
src/pulsecore/svolume_mmx.c
View file

@ -240,100 +240,10 @@ static void pa_volume_s16re_mmx(int16_t *samples, const int32_t *volumes, unsign
);
}
#undef RUN_TEST
#ifdef RUN_TEST
#define CHANNELS 2
#define SAMPLES 1022
#define TIMES 1000
#define TIMES2 100
#define PADDING 16
static void run_test(void) {
int16_t samples[SAMPLES];
int16_t samples_ref[SAMPLES];
int16_t samples_orig[SAMPLES];
int32_t volumes[CHANNELS + PADDING];
int i, j, padding;
pa_do_volume_func_t func;
pa_usec_t start, stop;
int k;
pa_usec_t min = INT_MAX, max = 0;
double s1 = 0, s2 = 0;
func = pa_get_volume_func(PA_SAMPLE_S16NE);
printf("checking MMX %zd\n", sizeof(samples));
pa_random(samples, sizeof(samples));
/* for (i = 0; i < SAMPLES; i++)
samples[i] = -1; */
memcpy(samples_ref, samples, sizeof(samples));
memcpy(samples_orig, samples, sizeof(samples));
for (i = 0; i < CHANNELS; i++)
volumes[i] = PA_CLAMP_VOLUME(rand() >> 15);
/* volumes[i] = 0x0000ffff; */
for (padding = 0; padding < PADDING; padding++, i++)
volumes[i] = volumes[padding];
func(samples_ref, volumes, CHANNELS, sizeof(samples));
pa_volume_s16ne_mmx(samples, volumes, CHANNELS, sizeof(samples));
for (i = 0; i < SAMPLES; i++) {
if (samples[i] != samples_ref[i]) {
printf("%d: %04x != %04x (%04x * %08x)\n", i, samples[i], samples_ref[i],
samples_orig[i], volumes[i % CHANNELS]);
}
}
for (k = 0; k < TIMES2; k++) {
start = pa_rtclock_now();
for (j = 0; j < TIMES; j++) {
memcpy(samples, samples_orig, sizeof(samples));
pa_volume_s16ne_mmx(samples, volumes, CHANNELS, sizeof(samples));
}
stop = pa_rtclock_now();
if (min > (stop - start)) min = stop - start;
if (max < (stop - start)) max = stop - start;
s1 += stop - start;
s2 += (stop - start) * (stop - start);
}
pa_log_info("MMX: %llu usec (min = %llu, max = %llu, stddev = %g).", (long long unsigned int)s1,
(long long unsigned int)min, (long long unsigned int)max, sqrt(TIMES2 * s2 - s1 * s1) / TIMES2);
min = INT_MAX; max = 0;
s1 = s2 = 0;
for (k = 0; k < TIMES2; k++) {
start = pa_rtclock_now();
for (j = 0; j < TIMES; j++) {
memcpy(samples_ref, samples_orig, sizeof(samples));
func(samples_ref, volumes, CHANNELS, sizeof(samples));
}
stop = pa_rtclock_now();
if (min > (stop - start)) min = stop - start;
if (max < (stop - start)) max = stop - start;
s1 += stop - start;
s2 += (stop - start) * (stop - start);
}
pa_log_info("ref: %llu usec (min = %llu, max = %llu, stddev = %g).", (long long unsigned int)s1,
(long long unsigned int)min, (long long unsigned int)max, sqrt(TIMES2 * s2 - s1 * s1) / TIMES2);
pa_assert_se(memcmp(samples_ref, samples, sizeof(samples)) == 0);
}
#endif
#endif /* defined (__i386__) || defined (__amd64__) */
void pa_volume_func_init_mmx(pa_cpu_x86_flag_t flags) {
#if defined (__i386__) || defined (__amd64__)
#ifdef RUN_TEST
run_test();
#endif
if ((flags & PA_CPU_X86_MMX) && (flags & PA_CPU_X86_CMOV)) {
pa_log_info("Initialising MMX optimized volume functions.");

85
src/pulsecore/svolume_sse.c
View file

@ -252,95 +252,10 @@ static void pa_volume_s16re_sse2(int16_t *samples, const int32_t *volumes, unsig
);
}
#undef RUN_TEST
#ifdef RUN_TEST
#define CHANNELS 2
#define SAMPLES 1022
#define TIMES 1000
#define TIMES2 100
#define PADDING 16
static void run_test(void) {
int16_t samples[SAMPLES];
int16_t samples_ref[SAMPLES];
int16_t samples_orig[SAMPLES];
int32_t volumes[CHANNELS + PADDING];
int i, j, padding;
pa_do_volume_func_t func;
pa_usec_t start, stop;
int k;
pa_usec_t min = INT_MAX, max = 0;
double s1 = 0, s2 = 0;
func = pa_get_volume_func(PA_SAMPLE_S16NE);
printf("checking SSE2 %zd\n", sizeof(samples));
pa_random(samples, sizeof(samples));
memcpy(samples_ref, samples, sizeof(samples));
memcpy(samples_orig, samples, sizeof(samples));
for (i = 0; i < CHANNELS; i++)
volumes[i] = PA_CLAMP_VOLUME(rand() >> 15);
for (padding = 0; padding < PADDING; padding++, i++)
volumes[i] = volumes[padding];
func(samples_ref, volumes, CHANNELS, sizeof(samples));
pa_volume_s16ne_sse2(samples, volumes, CHANNELS, sizeof(samples));
for (i = 0; i < SAMPLES; i++) {
if (samples[i] != samples_ref[i]) {
printf ("%d: %04x != %04x (%04x * %04x)\n", i, samples[i], samples_ref[i],
samples_orig[i], volumes[i % CHANNELS]);
}
}
for (k = 0; k < TIMES2; k++) {
start = pa_rtclock_now();
for (j = 0; j < TIMES; j++) {
memcpy(samples, samples_orig, sizeof(samples));
pa_volume_s16ne_sse2(samples, volumes, CHANNELS, sizeof(samples));
}
stop = pa_rtclock_now();
if (min > (stop - start)) min = stop - start;
if (max < (stop - start)) max = stop - start;
s1 += stop - start;
s2 += (stop - start) * (stop - start);
}
pa_log_info("SSE: %llu usec (min = %llu, max = %llu, stddev = %g).", (long long unsigned int)s1,
(long long unsigned int)min, (long long unsigned int)max, sqrt(TIMES2 * s2 - s1 * s1) / TIMES2);
min = INT_MAX; max = 0;
s1 = s2 = 0;
for (k = 0; k < TIMES2; k++) {
start = pa_rtclock_now();
for (j = 0; j < TIMES; j++) {
memcpy(samples_ref, samples_orig, sizeof(samples));
func(samples_ref, volumes, CHANNELS, sizeof(samples));
}
stop = pa_rtclock_now();
if (min > (stop - start)) min = stop - start;
if (max < (stop - start)) max = stop - start;
s1 += stop - start;
s2 += (stop - start) * (stop - start);
}
pa_log_info("ref: %llu usec (min = %llu, max = %llu, stddev = %g).", (long long unsigned int)s1,
(long long unsigned int)min, (long long unsigned int)max, sqrt(TIMES2 * s2 - s1 * s1) / TIMES2);
pa_assert_se(memcmp(samples_ref, samples, sizeof(samples)) == 0);
}
#endif
#endif /* defined (__i386__) || defined (__amd64__) */
void pa_volume_func_init_sse(pa_cpu_x86_flag_t flags) {
#if defined (__i386__) || defined (__amd64__)
#ifdef RUN_TEST
run_test();
#endif
if (flags & PA_CPU_X86_SSE2) {
pa_log_info("Initialising SSE2 optimized volume functions.");