#include "sample-util.h"
-#if defined (__i386__) || defined (__amd64__)
+#if (!defined(__FreeBSD__) && defined (__i386__)) || defined (__amd64__)
/* in s: 2 int16_t samples
* in v: 2 int32_t volumes, fixed point 16:16
* out s: contains scaled and clamped int16_t samples.
*
* We calculate the high 32 bits of a 32x16 multiply which we then
- * clamp to 16 bits. The calulcation is:
+ * clamp to 16 bits. The calculation is:
*
* vl = (v & 0xffff)
* vh = (v >> 16)
" por %%mm4, "#s1" \n\t" /* .. | l h | */ \
" por %%mm5, "#s2" \n\t"
-static void pa_volume_s16ne_mmx(int16_t *samples, int32_t *volumes, unsigned channels, unsigned length) {
+static void pa_volume_s16ne_mmx(int16_t *samples, const int32_t *volumes, unsigned channels, unsigned length) {
pa_reg_x86 channel, temp;
/* Channels must be at least 4, and always a multiple of the original number.
" test $1, %2 \n\t" /* check for odd samples */
" je 2f \n\t"
- " movd (%1, %3, 4), %%mm0 \n\t" /* | v0h | v0l | */
+ " movd (%q1, %3, 4), %%mm0 \n\t" /* | v0h | v0l | */
" movw (%0), %w4 \n\t" /* .. | p0 | */
" movd %4, %%mm1 \n\t"
VOLUME_32x16 (%%mm1, %%mm0)
" je 4f \n\t"
"3: \n\t" /* do samples in groups of 2 */
- " movq (%1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
+ " movq (%q1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
" movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
VOLUME_32x16 (%%mm1, %%mm0)
" movd %%mm0, (%0) \n\t" /* .. | p1*v1 | p0*v0 | */
" je 6f \n\t"
"5: \n\t" /* do samples in groups of 4 */
- " movq (%1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
- " movq 8(%1, %3, 4), %%mm2 \n\t" /* | v3h | v3l | v2h | v2l | */
+ " movq (%q1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
+ " movq 8(%q1, %3, 4), %%mm2 \n\t" /* | v3h | v3l | v2h | v2l | */
" movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
" movd 4(%0), %%mm3 \n\t" /* .. | p3 | p2 | */
VOLUME_32x16 (%%mm1, %%mm0)
: "+r" (samples), "+r" (volumes), "+r" (length), "=D" (channel), "=&r" (temp)
#if defined (__i386__)
- : "m" ((pa_reg_x86)channels)
+ : "m" (channels)
#else
: "r" ((pa_reg_x86)channels)
#endif
);
}
-static void pa_volume_s16re_mmx(int16_t *samples, int32_t *volumes, unsigned channels, unsigned length) {
+static void pa_volume_s16re_mmx(int16_t *samples, const int32_t *volumes, unsigned channels, unsigned length) {
pa_reg_x86 channel, temp;
/* Channels must be at least 4, and always a multiple of the original number.
" test $1, %2 \n\t" /* check for odd samples */
" je 2f \n\t"
- " movd (%1, %3, 4), %%mm0 \n\t" /* | v0h | v0l | */
+ " movd (%q1, %3, 4), %%mm0 \n\t" /* | v0h | v0l | */
" movw (%0), %w4 \n\t" /* .. | p0 | */
" rorw $8, %w4 \n\t"
" movd %4, %%mm1 \n\t"
" je 4f \n\t"
"3: \n\t" /* do samples in groups of 2 */
- " movq (%1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
+ " movq (%q1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
" movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
SWAP_16 (%%mm1)
VOLUME_32x16 (%%mm1, %%mm0)
" je 6f \n\t"
"5: \n\t" /* do samples in groups of 4 */
- " movq (%1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
- " movq 8(%1, %3, 4), %%mm2 \n\t" /* | v3h | v3l | v2h | v2l | */
+ " movq (%q1, %3, 4), %%mm0 \n\t" /* | v1h | v1l | v0h | v0l | */
+ " movq 8(%q1, %3, 4), %%mm2 \n\t" /* | v3h | v3l | v2h | v2l | */
" movd (%0), %%mm1 \n\t" /* .. | p1 | p0 | */
" movd 4(%0), %%mm3 \n\t" /* .. | p3 | p2 | */
SWAP_16_2 (%%mm1, %%mm3)
: "+r" (samples), "+r" (volumes), "+r" (length), "=D" (channel), "=&r" (temp)
#if defined (__i386__)
- : "m" ((pa_reg_x86)channels)
+ : "m" (channels)
#else
: "r" ((pa_reg_x86)channels)
#endif
);
}
-#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() >> 1);
- /* 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__) */
-
+#endif /* (!defined(__FreeBSD__) && 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 (!defined(__FreeBSD__) && defined (__i386__)) || defined (__amd64__)
if ((flags & PA_CPU_X86_MMX) && (flags & PA_CPU_X86_CMOV)) {
pa_log_info("Initialising MMX optimized volume functions.");
pa_set_volume_func(PA_SAMPLE_S16NE, (pa_do_volume_func_t) pa_volume_s16ne_mmx);
pa_set_volume_func(PA_SAMPLE_S16RE, (pa_do_volume_func_t) pa_volume_s16re_mmx);
}
-#endif /* defined (__i386__) || defined (__amd64__) */
+#endif /* (!defined(__FreeBSD__) && defined (__i386__)) || defined (__amd64__) */
}