tests: Add NEON mix test cases to cpu-test

on beagle-xm (Cortex-A8)

Initialising ARM NEON optimized mixing functions.
Checking NEON mix
Testing 2-channel mixing performance with 7 sample alignment
func: 2329073 usec (avg: 23290.7, min = 18127, max = 65368, stddev = 10404.2).
orig: 7931126 usec (avg: 79311.3, min = 65002, max = 239411, stddev = 35885.6).

Signed-off-by: Peter Meerwald <pmeerw@pmeerw.net>
This commit is contained in:
Peter Meerwald 2013-02-13 17:27:10 +01:00 committed by Tanu Kaskinen
parent 1e4e586150
commit cf8e67ede8

View file

@ -34,6 +34,7 @@
#include <pulsecore/sconv.h>
#include <pulsecore/remap.h>
#include <pulsecore/sample-util.h>
#include <pulsecore/mix.h>
#define PA_CPU_TEST_RUN_START(l, t1, t2) \
{ \
@ -698,6 +699,156 @@ END_TEST
#undef TIMES2
/* End remap tests */
/* Start mix tests */
/* Only ARM NEON has mix tests, so disable the related functions for other
* architectures for now to avoid compiler warnings about unused functions. */
#if defined (__arm__) && defined (__linux__)
#ifdef HAVE_NEON
#define SAMPLES 1028
#define TIMES 1000
#define TIMES2 100
static void acquire_mix_streams(pa_mix_info streams[], unsigned nstreams) {
unsigned i;
for (i = 0; i < nstreams; i++)
streams[i].ptr = pa_memblock_acquire_chunk(&streams[i].chunk);
}
static void release_mix_streams(pa_mix_info streams[], unsigned nstreams) {
unsigned i;
for (i = 0; i < nstreams; i++)
pa_memblock_release(streams[i].chunk.memblock);
}
static void run_mix_test(
pa_do_mix_func_t func,
pa_do_mix_func_t orig_func,
int align,
int channels,
pa_bool_t correct,
pa_bool_t perf) {
PA_DECLARE_ALIGNED(8, int16_t, in0[SAMPLES * 4]) = { 0 };
PA_DECLARE_ALIGNED(8, int16_t, in1[SAMPLES * 4]) = { 0 };
PA_DECLARE_ALIGNED(8, int16_t, out[SAMPLES * 4]) = { 0 };
PA_DECLARE_ALIGNED(8, int16_t, out_ref[SAMPLES * 4]) = { 0 };
int16_t *samples0, *samples1;
int16_t *samples, *samples_ref;
int nsamples;
pa_mempool *pool;
pa_memchunk c0, c1;
pa_mix_info m[2];
int i;
pa_assert(channels == 1 || channels == 2 || channels == 4);
/* Force sample alignment as requested */
samples0 = in0 + (8 - align);
samples1 = in1 + (8 - align);
samples = out + (8 - align);
samples_ref = out_ref + (8 - align);
nsamples = channels * (SAMPLES - (8 - align));
fail_unless((pool = pa_mempool_new(FALSE, 0)) != NULL, NULL);
pa_random(samples0, nsamples * sizeof(int16_t));
c0.memblock = pa_memblock_new_fixed(pool, samples0, nsamples * sizeof(int16_t), FALSE);
c0.length = pa_memblock_get_length(c0.memblock);
c0.index = 0;
pa_random(samples1, nsamples * sizeof(int16_t));
c1.memblock = pa_memblock_new_fixed(pool, samples1, nsamples * sizeof(int16_t), FALSE);
c1.length = pa_memblock_get_length(c1.memblock);
c1.index = 0;
m[0].chunk = c0;
m[0].volume.channels = channels;
for (i = 0; i < channels; i++) {
m[0].volume.values[i] = PA_VOLUME_NORM;
m[0].linear[i].i = 0x5555;
}
m[1].chunk = c1;
m[1].volume.channels = channels;
for (i = 0; i < channels; i++) {
m[1].volume.values[i] = PA_VOLUME_NORM;
m[1].linear[i].i = 0x6789;
}
if (correct) {
acquire_mix_streams(m, 2);
orig_func(m, 2, channels, samples_ref, nsamples * sizeof(int16_t));
release_mix_streams(m, 2);
acquire_mix_streams(m, 2);
func(m, 2, channels, samples, nsamples * sizeof(int16_t));
release_mix_streams(m, 2);
for (i = 0; i < nsamples; i++) {
if (samples[i] != samples_ref[i]) {
pa_log_debug("Correctness test failed: align=%d, channels=%d", align, channels);
pa_log_debug("%d: %hd != %04hd (%hd + %hd)\n",
i,
samples[i], samples_ref[i],
samples0[i], samples1[i]);
fail();
}
}
}
if (perf) {
pa_log_debug("Testing %d-channel mixing performance with %d sample alignment", channels, align);
PA_CPU_TEST_RUN_START("func", TIMES, TIMES2) {
acquire_mix_streams(m, 2);
func(m, 2, channels, samples, nsamples * sizeof(int16_t));
release_mix_streams(m, 2);
} PA_CPU_TEST_RUN_STOP
PA_CPU_TEST_RUN_START("orig", TIMES, TIMES2) {
acquire_mix_streams(m, 2);
orig_func(m, 2, channels, samples_ref, nsamples * sizeof(int16_t));
release_mix_streams(m, 2);
} PA_CPU_TEST_RUN_STOP
}
pa_memblock_unref(c0.memblock);
pa_memblock_unref(c1.memblock);
pa_mempool_free(pool);
}
#endif /* HAVE_NEON */
#endif /* defined (__arm__) && defined (__linux__) */
#if defined (__arm__) && defined (__linux__)
#ifdef HAVE_NEON
START_TEST (mix_neon_test) {
pa_do_mix_func_t orig_func, neon_func;
pa_cpu_arm_flag_t flags = 0;
pa_cpu_get_arm_flags(&flags);
if (!(flags & PA_CPU_ARM_NEON)) {
pa_log_info("NEON not supported. Skipping");
return;
}
orig_func = pa_get_mix_func(PA_SAMPLE_S16NE);
pa_mix_func_init_neon(flags);
neon_func = pa_get_mix_func(PA_SAMPLE_S16NE);
pa_log_debug("Checking NEON mix");
run_mix_test(neon_func, orig_func, 7, 2, TRUE, TRUE);
}
END_TEST
#endif /* HAVE_NEON */
#endif /* defined (__arm__) && defined (__linux__) */
/* End mix tests */
int main(int argc, char *argv[]) {
int failed = 0;
Suite *s;
@ -741,6 +892,15 @@ int main(int argc, char *argv[]) {
#if defined (__i386__) || defined (__amd64__)
tcase_add_test(tc, remap_mmx_test);
tcase_add_test(tc, remap_sse2_test);
#endif
tcase_set_timeout(tc, 120);
suite_add_tcase(s, tc);
/* Mix tests */
tc = tcase_create("mix");
#if defined (__arm__) && defined (__linux__)
#if HAVE_NEON
tcase_add_test(tc, mix_neon_test);
#endif
#endif
tcase_set_timeout(tc, 120);
suite_add_tcase(s, tc);