src/pulsecore/cpu-arm.c | 12 +++- src/pulsecore/cpu-arm.h | 1 src/pulsecore/svolume_arm.c | 121 ++++++++++---------------------------------- src/tests/cpu-test.c | 43 +++++++++++++++ 4 files changed, 81 insertions(+), 96 deletions(-)
New commits: commit 97f2d6b347982eb2ffa9b1a14ae5ffd8a875f423 Author: Arun Raghavan <[email protected]> Date: Thu Oct 18 10:43:17 2012 +0530 core: Fix a litte-endian bug in ARM svolume code The old code seems to be written for big-endian code. diff --git a/src/pulsecore/svolume_arm.c b/src/pulsecore/svolume_arm.c index b5ccc67..f06a702 100644 --- a/src/pulsecore/svolume_arm.c +++ b/src/pulsecore/svolume_arm.c @@ -75,13 +75,22 @@ static void pa_volume_s16ne_arm(int16_t *samples, const int32_t *volumes, unsign " ldrd r2, [r6], #8 \n\t" /* 2 samples at a time */ " ldr r0, [%0] \n\t" +#ifdef WORDS_BIGENDIAN " smulwt r2, r2, r0 \n\t" " smulwb r3, r3, r0 \n\t" +#else + " smulwb r2, r2, r0 \n\t" + " smulwt r3, r3, r0 \n\t" +#endif " ssat r2, #16, r2 \n\t" " ssat r3, #16, r3 \n\t" +#ifdef WORDS_BIGENDIAN " pkhbt r0, r3, r2, LSL #16 \n\t" +#else + " pkhbt r0, r2, r3, LSL #16 \n\t" +#endif " str r0, [%0], #4 \n\t" MOD_INC() @@ -95,18 +104,30 @@ static void pa_volume_s16ne_arm(int16_t *samples, const int32_t *volumes, unsign " ldrd r4, [r6], #8 \n\t" " ldrd r0, [%0] \n\t" +#ifdef WORDS_BIGENDIAN " smulwt r2, r2, r0 \n\t" " smulwb r3, r3, r0 \n\t" " smulwt r4, r4, r1 \n\t" " smulwb r5, r5, r1 \n\t" +#else + " smulwb r2, r2, r0 \n\t" + " smulwt r3, r3, r0 \n\t" + " smulwb r4, r4, r1 \n\t" + " smulwt r5, r5, r1 \n\t" +#endif " ssat r2, #16, r2 \n\t" " ssat r3, #16, r3 \n\t" " ssat r4, #16, r4 \n\t" " ssat r5, #16, r5 \n\t" +#ifdef WORDS_BIGENDIAN " pkhbt r0, r3, r2, LSL #16 \n\t" " pkhbt r1, r5, r4, LSL #16 \n\t" +#else + " pkhbt r0, r2, r3, LSL #16 \n\t" + " pkhbt r1, r4, r5, LSL #16 \n\t" +#endif " strd r0, [%0], #8 \n\t" MOD_INC() commit 600e3daa56394364f6f2497d55bbaea8bee0bec9 Author: Arun Raghavan <[email protected]> Date: Thu Oct 18 10:42:48 2012 +0530 core: Document ARM-optimised svolume code a bit diff --git a/src/pulsecore/svolume_arm.c b/src/pulsecore/svolume_arm.c index 92b05a4..b5ccc67 100644 --- a/src/pulsecore/svolume_arm.c +++ b/src/pulsecore/svolume_arm.c @@ -50,19 +50,19 @@ static void pa_volume_s16ne_arm(int16_t *samples, const int32_t *volumes, unsign ve = volumes + channels; __asm__ __volatile__ ( - " mov r6, %1 \n\t" + " mov r6, %1 \n\t" /* r6 = volumes */ " mov %3, %3, LSR #1 \n\t" /* length /= sizeof (int16_t) */ " tst %3, #1 \n\t" /* check for odd samples */ " beq 2f \n\t" - "1: \n\t" - " ldr r0, [r6], #4 \n\t" /* odd samples volumes */ - " ldrh r2, [%0] \n\t" + "1: \n\t" /* odd samples volumes */ + " ldr r0, [r6], #4 \n\t" /* r0 = volume */ + " ldrh r2, [%0] \n\t" /* r2 = sample */ - " smulwb r0, r0, r2 \n\t" - " ssat r0, #16, r0 \n\t" + " smulwb r0, r0, r2 \n\t" /* r0 = (r0 * r2) >> 16 */ + " ssat r0, #16, r0 \n\t" /* r0 = PA_CLAMP(r0, 0x7FFF) */ - " strh r0, [%0], #2 \n\t" + " strh r0, [%0], #2 \n\t" /* sample = r0 */ MOD_INC() commit b87067a007b7a957bd5c30eb0ef24d380c41c06f Author: Arun Raghavan <[email protected]> Date: Fri Oct 19 18:43:17 2012 +0530 tests: Minor cpu-test reorganisation Make volume and conversion tests separate test cases. diff --git a/src/tests/cpu-test.c b/src/tests/cpu-test.c index e23bfe5..e8ece7f 100644 --- a/src/tests/cpu-test.c +++ b/src/tests/cpu-test.c @@ -261,9 +261,9 @@ int main(int argc, char *argv[]) { pa_log_set_level(PA_LOG_DEBUG); s = suite_create("CPU"); - tc = tcase_create("x86"); /* Volume tests */ + tc = tcase_create("svolume"); #if defined (__i386__) || defined (__amd64__) tcase_add_test(tc, svolume_mmx_test); tcase_add_test(tc, svolume_sse_test); @@ -272,8 +272,10 @@ int main(int argc, char *argv[]) { tcase_add_test(tc, svolume_arm_test); #endif tcase_add_test(tc, svolume_orc_test); + suite_add_tcase(s, tc); /* Converstion tests */ + tc = tcase_create("sconv"); #if defined (__i386__) || defined (__amd64__) tcase_add_test(tc, sconv_sse_test); #endif commit f8017bb19d6eb815b2dafed521f7a0447b6d5501 Author: Arun Raghavan <[email protected]> Date: Tue Oct 16 12:23:19 2012 +0530 tests: Factor out ARM svolume test into cpu-test This allows us to run the testing code separately from the PulseAudio daemon, which makes developing / regression testing this code a lot easier. diff --git a/src/pulsecore/svolume_arm.c b/src/pulsecore/svolume_arm.c index 79ed830..92b05a4 100644 --- a/src/pulsecore/svolume_arm.c +++ b/src/pulsecore/svolume_arm.c @@ -121,98 +121,12 @@ static void pa_volume_s16ne_arm(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 ARM %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_arm(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_arm(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("ARM: %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 (__arm__) && defined (HAVE_ARMV6) */ - void pa_volume_func_init_arm(pa_cpu_arm_flag_t flags) { #if defined (__arm__) && defined (HAVE_ARMV6) pa_log_info("Initialising ARM optimized volume functions."); -#ifdef RUN_TEST - run_test(); -#endif - pa_set_volume_func(PA_SAMPLE_S16NE, (pa_do_volume_func_t) pa_volume_s16ne_arm); #endif /* defined (__arm__) && defined (HAVE_ARMV6) */ } diff --git a/src/tests/cpu-test.c b/src/tests/cpu-test.c index ac4312e..e23bfe5 100644 --- a/src/tests/cpu-test.c +++ b/src/tests/cpu-test.c @@ -90,6 +90,7 @@ static void run_volume_test(pa_do_volume_func_t func, pa_do_volume_func_t orig_f fail_unless(memcmp(samples_ref, samples, sizeof(samples)) == 0); } +#if defined (__i386__) || defined (__amd64__) START_TEST (svolume_mmx_test) { pa_do_volume_func_t orig_func, mmx_func; pa_cpu_x86_flag_t flags = 0; @@ -129,6 +130,29 @@ START_TEST (svolume_sse_test) { run_volume_test(sse_func, orig_func); } END_TEST +#endif /* defined (__i386__) || defined (__amd64__) */ + +#if defined (__arm__) && defined (__linux__) +START_TEST (svolume_arm_test) { + pa_do_volume_func_t orig_func, arm_func; + pa_cpu_arm_flag_t flags = 0; + + pa_cpu_get_arm_flags(&flags); + + if (!(flags & PA_CPU_ARM_V6)) { + pa_log_info("ARMv6 instructions not supported. Skipping"); + return; + } + + orig_func = pa_get_volume_func(PA_SAMPLE_S16NE); + pa_volume_func_init_arm(flags); + arm_func = pa_get_volume_func(PA_SAMPLE_S16NE); + + pa_log_debug("Checking ARM svolume"); + run_volume_test(arm_func, orig_func); +} +END_TEST +#endif /* defined (__arm__) && defined (__linux__) */ START_TEST (svolume_orc_test) { pa_do_volume_func_t orig_func, orc_func; @@ -161,6 +185,8 @@ END_TEST #undef PADDING /* End svolume tests */ +/* Start conversion tests */ +#if defined (__i386__) || defined (__amd64__) START_TEST (sconv_sse_test) { #define SAMPLES 1019 #define TIMES 1000 @@ -222,6 +248,8 @@ START_TEST (sconv_sse_test) { #undef TIMES } END_TEST +#endif /* defined (__i386__) || defined (__amd64__) */ +/* End conversion tests */ int main(int argc, char *argv[]) { int failed = 0; @@ -234,10 +262,21 @@ int main(int argc, char *argv[]) { s = suite_create("CPU"); tc = tcase_create("x86"); + + /* Volume tests */ +#if defined (__i386__) || defined (__amd64__) tcase_add_test(tc, svolume_mmx_test); tcase_add_test(tc, svolume_sse_test); +#endif +#if defined (__arm__) && defined (__linux__) + tcase_add_test(tc, svolume_arm_test); +#endif tcase_add_test(tc, svolume_orc_test); + + /* Converstion tests */ +#if defined (__i386__) || defined (__amd64__) tcase_add_test(tc, sconv_sse_test); +#endif suite_add_tcase(s, tc); sr = srunner_create(s); commit 60660a1aecadb48f01e0e52fe0faf8e450d1e865 Author: Arun Raghavan <[email protected]> Date: Tue Oct 16 12:00:19 2012 +0530 core: Separate ARM CPU detection from initialisation This separate the ARM CPU feature detection code from the initialisation so that we can reuse it in tests, like we did for x86 a while back. diff --git a/src/pulsecore/cpu-arm.c b/src/pulsecore/cpu-arm.c index a724309..934c7fe 100644 --- a/src/pulsecore/cpu-arm.c +++ b/src/pulsecore/cpu-arm.c @@ -79,7 +79,7 @@ static char *get_cpuinfo(void) { } #endif /* defined (__arm__) && defined (__linux__) */ -pa_bool_t pa_cpu_init_arm(pa_cpu_arm_flag_t *flags) { +void pa_cpu_get_arm_flags(pa_cpu_arm_flag_t *flags) { #if defined (__arm__) #if defined (__linux__) char *cpuinfo, *line; @@ -89,7 +89,7 @@ pa_bool_t pa_cpu_init_arm(pa_cpu_arm_flag_t *flags) { * space support to get the CPU features. This only works on linux AFAIK. */ if (!(cpuinfo = get_cpuinfo())) { pa_log("Can't read cpuinfo"); - return FALSE; + return; } *flags = 0; @@ -131,6 +131,14 @@ pa_bool_t pa_cpu_init_arm(pa_cpu_arm_flag_t *flags) { (*flags & PA_CPU_ARM_EDSP) ? "EDSP " : "", (*flags & PA_CPU_ARM_NEON) ? "NEON " : "", (*flags & PA_CPU_ARM_VFPV3) ? "VFPV3 " : ""); +#endif +#endif +} + +pa_bool_t pa_cpu_init_arm(pa_cpu_arm_flag_t *flags) { +#if defined (__arm__) +#if defined (__linux__) + pa_cpu_get_arm_flags(flags); if (*flags & PA_CPU_ARM_V6) pa_volume_func_init_arm(*flags); diff --git a/src/pulsecore/cpu-arm.h b/src/pulsecore/cpu-arm.h index f0f49ef..91aa659 100644 --- a/src/pulsecore/cpu-arm.h +++ b/src/pulsecore/cpu-arm.h @@ -35,6 +35,7 @@ typedef enum pa_cpu_arm_flag { PA_CPU_ARM_VFPV3 = (1 << 5) } pa_cpu_arm_flag_t; +void pa_cpu_get_arm_flags(pa_cpu_arm_flag_t *flags); pa_bool_t pa_cpu_init_arm(pa_cpu_arm_flag_t *flags); /* some optimized functions */ _______________________________________________ pulseaudio-commits mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/pulseaudio-commits
