Merge branch 'master' of git://0pointer.de/pulseaudio into dbus-work

Conflicts:
	src/Makefile.am

69 files changed, 3739 insertions, 1052 deletions

diff --git a/src/.gitignore b/src/.gitignore
index 82331524..6cd173c0 100644
--- a/src/.gitignore
+++ b/src/.gitignore
@@ -1,3 +1,4 @@
+usergroup-test
 sigbus-test
 TAGS
 alsa-time-test
diff --git a/src/Makefile.am b/src/Makefile.am
index a46c5a1b..c3ca1eb1 100644
--- a/src/Makefile.am
+++ b/src/Makefile.am
@@ -74,6 +74,7 @@ AM_CFLAGS = \
 	$(LIBSAMPLERATE_CFLAGS) \
 	$(LIBSNDFILE_CFLAGS) \
 	$(LIBSPEEX_CFLAGS) \
+	-DPA_BUILDDIR=\"$(abs_builddir)\" \
 	-DPA_DLSEARCHPATH=\"$(modlibexecdir)\" \
 	-DPA_DEFAULT_CONFIG_DIR=\"$(PA_DEFAULT_CONFIG_DIR)\" \
 	-DPA_BINARY=\"$(PA_BINARY)\" \
@@ -83,8 +84,8 @@ AM_CFLAGS = \
 	-DAO_REQUIRE_CAS \
 	-DPULSE_LOCALEDIR=\"$(pulselocaledir)\" \
 	-DPA_MACHINE_ID=\"$(localstatedir)/lib/dbus/machine-id\" \
-	-DPA_ALSA_PATHS_DIR=\"$(alsapathsdir)\" \
-	-DPA_ALSA_PROFILE_SETS_DIR=\"$(alsaprofilesetsdir)\"
+        -DPA_ALSA_PATHS_DIR=\"$(alsapathsdir)\" \
+        -DPA_ALSA_PROFILE_SETS_DIR=\"$(alsaprofilesetsdir)\"
 
 AM_LIBADD = $(PTHREAD_LIBS) $(INTLLIBS)
 AM_LDADD = $(PTHREAD_LIBS) $(INTLLIBS)
@@ -280,7 +281,8 @@ TESTS = \
 		proplist-test \
 		lock-autospawn-test \
 		prioq-test \
-		sigbus-test
+		sigbus-test \
+		usergroup-test
 
 TESTS_BINARIES = \
 		mainloop-test \
@@ -318,7 +320,8 @@ TESTS_BINARIES = \
 		stripnul \
 		lock-autospawn-test \
 		prioq-test \
-		sigbus-test
+		sigbus-test \
+		usergroup-test
 
 if HAVE_SIGXCPU
 #TESTS += \
@@ -557,6 +560,11 @@ alsa_time_test_LDADD = $(AM_LDADD)
 alsa_time_test_CFLAGS = $(AM_CFLAGS) $(ASOUNDLIB_CFLAGS)
 alsa_time_test_LDFLAGS = $(AM_LDFLAGS) $(BINLDFLAGS) $(ASOUNDLIB_LIBS)
 
+usergroup_test_SOURCES = tests/usergroup-test.c
+usergroup_test_LDADD = $(AM_LDADD) libpulsecore-@PA_MAJORMINORMICRO@.la
+usergroup_test_CFLAGS = $(AM_CFLAGS)
+usergroup_test_LDFLAGS = $(AM_LDFLAGS) $(BINLDFLAGS)
+
 ###################################
 #         Common library          #
 ###################################
@@ -621,6 +629,7 @@ libpulsecommon_@PA_MAJORMINORMICRO@_la_SOURCES = \
 		pulsecore/tagstruct.c pulsecore/tagstruct.h \
 		pulsecore/time-smoother.c pulsecore/time-smoother.h \
 		pulsecore/tokenizer.c pulsecore/tokenizer.h \
+		pulsecore/usergroup.c pulsecore/usergroup.h \
 		pulsecore/sndfile-util.c pulsecore/sndfile-util.h \
 		pulsecore/winsock.h
 
@@ -823,10 +832,14 @@ libpulsecore_@PA_MAJORMINORMICRO@_la_SOURCES = \
 		pulsecore/object.c pulsecore/object.h \
 		pulsecore/play-memblockq.c pulsecore/play-memblockq.h \
 		pulsecore/play-memchunk.c pulsecore/play-memchunk.h \
-		pulsecore/protocol-dbus.h \
+		pulsecore/remap.c pulsecore/remap.h \
+		pulsecore/remap_mmx.c \
 		pulsecore/resampler.c pulsecore/resampler.h \
 		pulsecore/rtpoll.c pulsecore/rtpoll.h \
 		pulsecore/sample-util.c pulsecore/sample-util.h \
+		pulsecore/cpu-arm.c pulsecore/cpu-x86.c \
+		pulsecore/svolume_c.c pulsecore/svolume_arm.c\
+		pulsecore/svolume_mmx.c pulsecore/svolume_sse.c \
 		pulsecore/sconv-s16be.c pulsecore/sconv-s16be.h \
 		pulsecore/sconv-s16le.c pulsecore/sconv-s16le.h \
 		pulsecore/sconv.c pulsecore/sconv.h \
diff --git a/src/daemon/cmdline.c b/src/daemon/cmdline.c
index ecb38486..f6cdcdc8 100644
--- a/src/daemon/cmdline.c
+++ b/src/daemon/cmdline.c
@@ -385,11 +385,6 @@ int pa_cmdline_parse(pa_daemon_conf *conf, int argc, char *const argv [], int *d
     pa_xfree(conf->script_commands);
     conf->script_commands = pa_strbuf_tostring_free(buf);
 
-    if (!conf->script_commands) {
-        pa_xfree(conf->script_commands);
-        conf->script_commands = NULL;
-    }
-
     *d = optind;
 
     return 0;
diff --git a/src/daemon/daemon-conf.c b/src/daemon/daemon-conf.c
index 31f29009..3428f807 100644
--- a/src/daemon/daemon-conf.c
+++ b/src/daemon/daemon-conf.c
@@ -136,9 +136,25 @@ static const pa_daemon_conf default_conf = {
 };
 
 pa_daemon_conf* pa_daemon_conf_new(void) {
-    pa_daemon_conf *c = pa_xnewdup(pa_daemon_conf, &default_conf, 1);
+    pa_daemon_conf *c;
+
+    c = pa_xnewdup(pa_daemon_conf, &default_conf, 1);
+
+#if defined(__linux__) && !defined(__OPTIMIZE__)
+
+    /* We abuse __OPTIMIZE__ as a check whether we are a debug build
+     * or not. If we are and are run from the build tree then we
+     * override the search path to point to our build tree */
+
+    if (pa_run_from_build_tree()) {
+        pa_log_notice("Detected that we are run from the build tree, fixing search path.");
+        c->dl_search_path = pa_xstrdup(PA_BUILDDIR "/.libs/");
+
+    } else
+
+#endif
+        c->dl_search_path = pa_xstrdup(PA_DLSEARCHPATH);
 
-    c->dl_search_path = pa_xstrdup(PA_DLSEARCHPATH);
     return c;
 }
 
diff --git a/src/daemon/main.c b/src/daemon/main.c
index 73696005..409823b1 100644
--- a/src/daemon/main.c
+++ b/src/daemon/main.c
@@ -39,8 +39,6 @@
 #include <sys/types.h>
 #include <sys/stat.h>
 
-#include <liboil/liboil.h>
-
 #ifdef HAVE_SYS_MMAN_H
 #include <sys/mman.h>
 #endif
@@ -95,6 +93,8 @@
 #ifdef HAVE_DBUS
 #include <pulsecore/dbus-shared.h>
 #endif
+#include <pulsecore/cpu-arm.h>
+#include <pulsecore/cpu-x86.h>
 
 #include "cmdline.h"
 #include "cpulimit.h"
@@ -804,6 +804,8 @@ int main(int argc, char *argv[]) {
     pa_log_info(_("Using state directory %s."), s);
     pa_xfree(s);
 
+    pa_log_info(_("Using modules directory %s."), conf->dl_search_path);
+
     pa_log_info(_("Running in system mode: %s"), pa_yes_no(pa_in_system_mode()));
 
     if (pa_in_system_mode())
@@ -851,6 +853,9 @@ int main(int argc, char *argv[]) {
 
     pa_memtrap_install();
 
+    pa_cpu_init_x86();
+    pa_cpu_init_arm();
+
     pa_assert_se(mainloop = pa_mainloop_new());
 
     if (!(c = pa_core_new(pa_mainloop_get_api(mainloop), !conf->disable_shm, conf->shm_size))) {
@@ -893,8 +898,6 @@ int main(int argc, char *argv[]) {
     win32_timer = pa_mainloop_get_api(mainloop)->rtclock_time_new(pa_mainloop_get_api(mainloop), pa_gettimeofday(&win32_tv), message_cb, NULL);
 #endif
 
-    oil_init();
-
     if (!conf->no_cpu_limit)
         pa_assert_se(pa_cpu_limit_init(pa_mainloop_get_api(mainloop)) == 0);
 
diff --git a/src/modules/alsa/alsa-mixer.c b/src/modules/alsa/alsa-mixer.c
index a4c2ee0f..61c92cd0 100644
--- a/src/modules/alsa/alsa-mixer.c
+++ b/src/modules/alsa/alsa-mixer.c
@@ -929,7 +929,7 @@ static int element_zero_volume(pa_alsa_element *e, snd_mixer_t *m) {
 
 int pa_alsa_path_select(pa_alsa_path *p, snd_mixer_t *m) {
     pa_alsa_element *e;
-    int r;
+    int r = 0;
 
     pa_assert(m);
     pa_assert(p);
@@ -1849,7 +1849,12 @@ pa_alsa_path* pa_alsa_path_new(const char *fname, pa_alsa_direction_t direction)
     items[1].data = &p->description;
     items[2].data = &p->name;
 
-    fn = pa_maybe_prefix_path(fname, PA_ALSA_PATHS_DIR);
+    fn = pa_maybe_prefix_path(fname,
+#if defined(__linux__) && !defined(__OPTIMIZE__)
+                              pa_run_from_build_tree() ? PA_BUILDDIR "/modules/alsa/mixer/paths/" :
+#endif
+                              PA_ALSA_PATHS_DIR);
+
     r = pa_config_parse(fn, NULL, items, p);
     pa_xfree(fn);
 
@@ -3110,7 +3115,12 @@ pa_alsa_profile_set* pa_alsa_profile_set_new(const char *fname, const pa_channel
     if (!fname)
         fname = "default.conf";
 
-    fn = pa_maybe_prefix_path(fname, PA_ALSA_PROFILE_SETS_DIR);
+    fn = pa_maybe_prefix_path(fname,
+#if defined(__linux__) && !defined(__OPTIMIZE__)
+                              pa_run_from_build_tree() ? PA_BUILDDIR "/modules/alsa/mixer/profile-sets/" :
+#endif
+                              PA_ALSA_PROFILE_SETS_DIR);
+
     r = pa_config_parse(fn, NULL, items, ps);
     pa_xfree(fn);
 
diff --git a/src/modules/alsa/alsa-sink.c b/src/modules/alsa/alsa-sink.c
index e3707ae7..07d53880 100644
--- a/src/modules/alsa/alsa-sink.c
+++ b/src/modules/alsa/alsa-sink.c
@@ -62,11 +62,21 @@
 /* #define DEBUG_TIMING */
 
 #define DEFAULT_DEVICE "default"
-#define DEFAULT_TSCHED_BUFFER_USEC (2*PA_USEC_PER_SEC)            /* 2s   -- Overall buffer size */
-#define DEFAULT_TSCHED_WATERMARK_USEC (20*PA_USEC_PER_MSEC)       /* 20ms -- Fill up when only this much is left in the buffer */
-#define TSCHED_WATERMARK_STEP_USEC (10*PA_USEC_PER_MSEC)          /* 10ms -- On underrun, increase watermark by this */
-#define TSCHED_MIN_SLEEP_USEC (10*PA_USEC_PER_MSEC)               /* 10ms -- Sleep at least 10ms on each iteration */
-#define TSCHED_MIN_WAKEUP_USEC (4*PA_USEC_PER_MSEC)               /* 4ms  -- Wakeup at least this long before the buffer runs empty*/
+
+#define DEFAULT_TSCHED_BUFFER_USEC (2*PA_USEC_PER_SEC)             /* 2s    -- Overall buffer size */
+#define DEFAULT_TSCHED_WATERMARK_USEC (20*PA_USEC_PER_MSEC)        /* 20ms  -- Fill up when only this much is left in the buffer */
+
+#define TSCHED_WATERMARK_INC_STEP_USEC (10*PA_USEC_PER_MSEC)       /* 10ms  -- On underrun, increase watermark by this */
+#define TSCHED_WATERMARK_DEC_STEP_USEC (5*PA_USEC_PER_MSEC)        /* 5ms   -- When everything's great, decrease watermark by this */
+#define TSCHED_WATERMARK_VERIFY_AFTER_USEC (20*PA_USEC_PER_SEC)    /* 20s   -- How long after a drop out recheck if things are good now */
+#define TSCHED_WATERMARK_INC_THRESHOLD_USEC (1*PA_USEC_PER_MSEC)   /* 3ms   -- If the buffer level ever below this theshold, increase the watermark */
+#define TSCHED_WATERMARK_DEC_THRESHOLD_USEC (100*PA_USEC_PER_MSEC) /* 100ms -- If the buffer level didn't drop below this theshold in the verification time, decrease the watermark */
+
+#define TSCHED_MIN_SLEEP_USEC (10*PA_USEC_PER_MSEC)                /* 10ms  -- Sleep at least 10ms on each iteration */
+#define TSCHED_MIN_WAKEUP_USEC (4*PA_USEC_PER_MSEC)                /* 4ms   -- Wakeup at least this long before the buffer runs empty*/
+
+#define SMOOTHER_MIN_INTERVAL (2*PA_USEC_PER_MSEC)                 /* 2ms   -- min smoother update interval */
+#define SMOOTHER_MAX_INTERVAL (200*PA_USEC_PER_MSEC)               /* 200ms -- max smoother update inteval */
 
 #define VOLUME_ACCURACY (PA_VOLUME_NORM/100)  /* don't require volume adjustments to be perfectly correct. don't necessarily extend granularity in software unless the differences get greater than this level */
 
@@ -96,7 +106,12 @@ struct userdata {
         hwbuf_unused,
         min_sleep,
         min_wakeup,
-        watermark_step;
+        watermark_inc_step,
+        watermark_dec_step,
+        watermark_inc_threshold,
+        watermark_dec_threshold;
+
+    pa_usec_t watermark_dec_not_before;
 
     unsigned nfragments;
     pa_memchunk memchunk;
@@ -115,6 +130,8 @@ struct userdata {
     pa_smoother *smoother;
     uint64_t write_count;
     uint64_t since_start;
+    pa_usec_t smoother_interval;
+    pa_usec_t last_smoother_update;
 
     pa_reserve_wrapper *reserve;
     pa_hook_slot *reserve_slot;
@@ -243,6 +260,7 @@ static void fix_min_sleep_wakeup(struct userdata *u) {
     size_t max_use, max_use_2;
 
     pa_assert(u);
+    pa_assert(u->use_tsched);
 
     max_use = u->hwbuf_size - u->hwbuf_unused;
     max_use_2 = pa_frame_align(max_use/2, &u->sink->sample_spec);
@@ -257,6 +275,7 @@ static void fix_min_sleep_wakeup(struct userdata *u) {
 static void fix_tsched_watermark(struct userdata *u) {
     size_t max_use;
     pa_assert(u);
+    pa_assert(u->use_tsched);
 
     max_use = u->hwbuf_size - u->hwbuf_unused;
 
@@ -267,7 +286,7 @@ static void fix_tsched_watermark(struct userdata *u) {
         u->tsched_watermark = u->min_wakeup;
 }
 
-static void adjust_after_underrun(struct userdata *u) {
+static void increase_watermark(struct userdata *u) {
     size_t old_watermark;
     pa_usec_t old_min_latency, new_min_latency;
 
@@ -276,31 +295,64 @@ static void adjust_after_underrun(struct userdata *u) {
 
     /* First, just try to increase the watermark */
     old_watermark = u->tsched_watermark;
-    u->tsched_watermark = PA_MIN(u->tsched_watermark * 2, u->tsched_watermark + u->watermark_step);
+    u->tsched_watermark = PA_MIN(u->tsched_watermark * 2, u->tsched_watermark + u->watermark_inc_step);
     fix_tsched_watermark(u);
 
     if (old_watermark != u->tsched_watermark) {
-        pa_log_notice("Increasing wakeup watermark to %0.2f ms",
-                      (double) pa_bytes_to_usec(u->tsched_watermark, &u->sink->sample_spec) / PA_USEC_PER_MSEC);
+        pa_log_info("Increasing wakeup watermark to %0.2f ms",
+                    (double) pa_bytes_to_usec(u->tsched_watermark, &u->sink->sample_spec) / PA_USEC_PER_MSEC);
         return;
     }
 
     /* Hmm, we cannot increase the watermark any further, hence let's raise the latency */
     old_min_latency = u->sink->thread_info.min_latency;
-    new_min_latency = PA_MIN(old_min_latency * 2, old_min_latency + TSCHED_WATERMARK_STEP_USEC);
+    new_min_latency = PA_MIN(old_min_latency * 2, old_min_latency + TSCHED_WATERMARK_INC_STEP_USEC);
     new_min_latency = PA_MIN(new_min_latency, u->sink->thread_info.max_latency);
 
     if (old_min_latency != new_min_latency) {
-        pa_log_notice("Increasing minimal latency to %0.2f ms",
-                      (double) new_min_latency / PA_USEC_PER_MSEC);
+        pa_log_info("Increasing minimal latency to %0.2f ms",
+                    (double) new_min_latency / PA_USEC_PER_MSEC);
 
         pa_sink_set_latency_range_within_thread(u->sink, new_min_latency, u->sink->thread_info.max_latency);
-        return;
     }
 
     /* When we reach this we're officialy fucked! */
 }
 
+static void decrease_watermark(struct userdata *u) {
+    size_t old_watermark;
+    pa_usec_t now;
+
+    pa_assert(u);
+    pa_assert(u->use_tsched);
+
+    now = pa_rtclock_now();
+
+    if (u->watermark_dec_not_before <= 0)
+        goto restart;
+
+    if (u->watermark_dec_not_before > now)
+        return;
+
+    old_watermark = u->tsched_watermark;
+
+    if (u->tsched_watermark < u->watermark_dec_step)
+        u->tsched_watermark = u->tsched_watermark / 2;
+    else
+        u->tsched_watermark = PA_MAX(u->tsched_watermark / 2, u->tsched_watermark - u->watermark_dec_step);
+
+    fix_tsched_watermark(u);
+
+    if (old_watermark != u->tsched_watermark)
+        pa_log_info("Decreasing wakeup watermark to %0.2f ms",
+                    (double) pa_bytes_to_usec(u->tsched_watermark, &u->sink->sample_spec) / PA_USEC_PER_MSEC);
+
+    /* We don't change the latency range*/
+
+restart:
+    u->watermark_dec_not_before = now + TSCHED_WATERMARK_VERIFY_AFTER_USEC;
+}
+
 static void hw_sleep_time(struct userdata *u, pa_usec_t *sleep_usec, pa_usec_t*process_usec) {
     pa_usec_t usec, wm;
 
@@ -308,6 +360,7 @@ static void hw_sleep_time(struct userdata *u, pa_usec_t *sleep_usec, pa_usec_t*p
     pa_assert(process_usec);
 
     pa_assert(u);
+    pa_assert(u->use_tsched);
 
     usec = pa_sink_get_requested_latency_within_thread(u->sink);
 
@@ -355,7 +408,7 @@ static int try_recover(struct userdata *u, const char *call, int err) {
     return 0;
 }
 
-static size_t check_left_to_play(struct userdata *u, size_t n_bytes) {
+static size_t check_left_to_play(struct userdata *u, size_t n_bytes, pa_bool_t on_timeout) {
     size_t left_to_play;
 
     /* We use <= instead of < for this check here because an underrun
@@ -363,34 +416,55 @@ static size_t check_left_to_play(struct userdata *u, size_t n_bytes) {
      * it is removed from the buffer. This is particularly important
      * when block transfer is used. */
 
-    if (n_bytes <= u->hwbuf_size) {
+    if (n_bytes <= u->hwbuf_size)
         left_to_play = u->hwbuf_size - n_bytes;
+    else {
+
+        /* We got a dropout. What a mess! */
+        left_to_play = 0;
 
 #ifdef DEBUG_TIMING
-        pa_log_debug("%0.2f ms left to play", (double) pa_bytes_to_usec(left_to_play, &u->sink->sample_spec) / PA_USEC_PER_MSEC);
+        PA_DEBUG_TRAP;
 #endif
 
-    } else {
-        left_to_play = 0;
+        if (!u->first && !u->after_rewind)
+            if (pa_log_ratelimit())
+                pa_log_info("Underrun!");
+    }
 
 #ifdef DEBUG_TIMING
-        PA_DEBUG_TRAP;
+    pa_log_debug("%0.2f ms left to play; inc threshold = %0.2f ms; dec threshold = %0.2f ms",
+                 (double) pa_bytes_to_usec(left_to_play, &u->sink->sample_spec) / PA_USEC_PER_MSEC,
+                 (double) pa_bytes_to_usec(u->watermark_inc_threshold, &u->sink->sample_spec) / PA_USEC_PER_MSEC,
+                 (double) pa_bytes_to_usec(u->watermark_dec_threshold, &u->sink->sample_spec) / PA_USEC_PER_MSEC);
 #endif
 
+    if (u->use_tsched) {
+        pa_bool_t reset_not_before = TRUE;
+
         if (!u->first && !u->after_rewind) {
+            if (left_to_play < u->watermark_inc_threshold)
+                increase_watermark(u);
+            else if (left_to_play > u->watermark_dec_threshold) {
+                reset_not_before = FALSE;
 
-            if (pa_log_ratelimit())
-                pa_log_info("Underrun!");
+                /* We decrease the watermark only if have actually
+                 * been woken up by a timeout. If something else woke
+                 * us up it's too easy to fulfill the deadlines... */
 
-            if (u->use_tsched)
-                adjust_after_underrun(u);
+                if (on_timeout)
+                    decrease_watermark(u);
+            }
         }
+
+        if (reset_not_before)
+            u->watermark_dec_not_before = 0;
     }
 
     return left_to_play;
 }
 
-static int mmap_write(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled) {
+static int mmap_write(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled, pa_bool_t on_timeout) {
     pa_bool_t work_done = TRUE;
     pa_usec_t max_sleep_usec = 0, process_usec = 0;
     size_t left_to_play;
@@ -425,7 +499,8 @@ static int mmap_write(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polle
         pa_log_debug("avail: %lu", (unsigned long) n_bytes);
 #endif
 
-        left_to_play = check_left_to_play(u, n_bytes);
+        left_to_play = check_left_to_play(u, n_bytes, on_timeout);
+        on_timeout = FALSE;
 
         if (u->use_tsched)
 
@@ -560,7 +635,7 @@ static int mmap_write(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polle
     return work_done ? 1 : 0;
 }
 
-static int unix_write(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled) {
+static int unix_write(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled, pa_bool_t on_timeout) {
     pa_bool_t work_done = FALSE;
     pa_usec_t max_sleep_usec = 0, process_usec = 0;
     size_t left_to_play;
@@ -586,7 +661,8 @@ static int unix_write(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polle
         }
 
         n_bytes = (size_t) n * u->frame_size;
-        left_to_play = check_left_to_play(u, n_bytes);
+        left_to_play = check_left_to_play(u, n_bytes, on_timeout);
+        on_timeout = FALSE;
 
         if (u->use_tsched)
 
@@ -723,18 +799,27 @@ static void update_smoother(struct userdata *u) {
         now1 = pa_timespec_load(&htstamp);
     }
 
+    /* Hmm, if the timestamp is 0, then it wasn't set and we take the current time */
+    if (now1 <= 0)
+        now1 = pa_rtclock_now();
+
+    /* check if the time since the last update is bigger than the interval */
+    if (u->last_smoother_update > 0)
+        if (u->last_smoother_update + u->smoother_interval > now1)
+            return;
+
     position = (int64_t) u->write_count - ((int64_t) delay * (int64_t) u->frame_size);
 
     if (PA_UNLIKELY(position < 0))
         position = 0;
 
-    /* Hmm, if the timestamp is 0, then it wasn't set and we take the current time */
-    if (now1 <= 0)
-        now1 = pa_rtclock_now();
-
     now2 = pa_bytes_to_usec((uint64_t) position, &u->sink->sample_spec);
 
     pa_smoother_put(u->smoother, now1, now2);
+
+    u->last_smoother_update = now1;
+    /* exponentially increase the update interval up to the MAX limit */
+    u->smoother_interval = PA_MIN (u->smoother_interval * 2, SMOOTHER_MAX_INTERVAL);
 }
 
 static pa_usec_t sink_get_latency(struct userdata *u) {
@@ -906,11 +991,12 @@ static int unsuspend(struct userdata *u) {
 
     u->write_count = 0;
     pa_smoother_reset(u->smoother, pa_rtclock_now(), TRUE);
+    u->smoother_interval = SMOOTHER_MIN_INTERVAL;
+    u->last_smoother_update = 0;
 
     u->first = TRUE;
     u->since_start = 0;
 
-
     pa_log_info("Resumed successfully...");
 
     return 0;
@@ -1263,15 +1349,16 @@ static void thread_func(void *userdata) {
         if (PA_SINK_IS_OPENED(u->sink->thread_info.state)) {
             int work_done;
             pa_usec_t sleep_usec = 0;
+            pa_bool_t on_timeout = pa_rtpoll_timer_elapsed(u->rtpoll);
 
             if (PA_UNLIKELY(u->sink->thread_info.rewind_requested))
                 if (process_rewind(u) < 0)
                         goto fail;
 
             if (u->use_mmap)
-                work_done = mmap_write(u, &sleep_usec, revents & POLLOUT);
+                work_done = mmap_write(u, &sleep_usec, revents & POLLOUT, on_timeout);
             else
-                work_done = unix_write(u, &sleep_usec, revents & POLLOUT);
+                work_done = unix_write(u, &sleep_usec, revents & POLLOUT, on_timeout);
 
             if (work_done < 0)
                 goto fail;
@@ -1622,6 +1709,7 @@ pa_sink *pa_alsa_sink_new(pa_module *m, pa_modargs *ma, const char*driver, pa_ca
             5,
             pa_rtclock_now(),
             TRUE);
+    u->smoother_interval = SMOOTHER_MIN_INTERVAL;
 
     dev_id = pa_modargs_get_value(
             ma, "device_id",
@@ -1771,7 +1859,6 @@ pa_sink *pa_alsa_sink_new(pa_module *m, pa_modargs *ma, const char*driver, pa_ca
     u->fragment_size = frag_size = (uint32_t) (period_frames * frame_size);
     u->nfragments = nfrags;
     u->hwbuf_size = u->fragment_size * nfrags;
-    u->tsched_watermark = pa_usec_to_bytes_round_up(pa_bytes_to_usec_round_up(tsched_watermark, &requested_ss), &u->sink->sample_spec);
     pa_cvolume_mute(&u->hardware_volume, u->sink->sample_spec.channels);
 
     pa_log_info("Using %u fragments of size %lu bytes, buffer time is %0.2fms",
@@ -1782,7 +1869,13 @@ pa_sink *pa_alsa_sink_new(pa_module *m, pa_modargs *ma, const char*driver, pa_ca
     pa_sink_set_max_rewind(u->sink, u->hwbuf_size);
 
     if (u->use_tsched) {
-        u->watermark_step = pa_usec_to_bytes(TSCHED_WATERMARK_STEP_USEC, &u->sink->sample_spec);
+        u->tsched_watermark = pa_usec_to_bytes_round_up(pa_bytes_to_usec_round_up(tsched_watermark, &requested_ss), &u->sink->sample_spec);
+
+        u->watermark_inc_step = pa_usec_to_bytes(TSCHED_WATERMARK_INC_STEP_USEC, &u->sink->sample_spec);
+        u->watermark_dec_step = pa_usec_to_bytes(TSCHED_WATERMARK_DEC_STEP_USEC, &u->sink->sample_spec);
+
+        u->watermark_inc_threshold = pa_usec_to_bytes_round_up(TSCHED_WATERMARK_INC_THRESHOLD_USEC, &u->sink->sample_spec);
+        u->watermark_dec_threshold = pa_usec_to_bytes_round_up(TSCHED_WATERMARK_DEC_THRESHOLD_USEC, &u->sink->sample_spec);
 
         fix_min_sleep_wakeup(u);
         fix_tsched_watermark(u);
@@ -1796,6 +1889,7 @@ pa_sink *pa_alsa_sink_new(pa_module *m, pa_modargs *ma, const char*driver, pa_ca
     } else
         pa_sink_set_fixed_latency(u->sink, pa_bytes_to_usec(u->hwbuf_size, &ss));
 
+
     reserve_update(u);
 
     if (update_sw_params(u) < 0)
diff --git a/src/modules/alsa/alsa-source.c b/src/modules/alsa/alsa-source.c
index 7da37553..165b2e3b 100644
--- a/src/modules/alsa/alsa-source.c
+++ b/src/modules/alsa/alsa-source.c
@@ -59,11 +59,22 @@
 /* #define DEBUG_TIMING */
 
 #define DEFAULT_DEVICE "default"
-#define DEFAULT_TSCHED_BUFFER_USEC (2*PA_USEC_PER_SEC)       /* 2s */
-#define DEFAULT_TSCHED_WATERMARK_USEC (20*PA_USEC_PER_MSEC)  /* 20ms */
-#define TSCHED_WATERMARK_STEP_USEC (10*PA_USEC_PER_MSEC)     /* 10ms */
-#define TSCHED_MIN_SLEEP_USEC (10*PA_USEC_PER_MSEC)          /* 10ms */
-#define TSCHED_MIN_WAKEUP_USEC (4*PA_USEC_PER_MSEC)          /* 4ms */
+
+#define DEFAULT_TSCHED_BUFFER_USEC (2*PA_USEC_PER_SEC)             /* 2s */
+#define DEFAULT_TSCHED_WATERMARK_USEC (20*PA_USEC_PER_MSEC)        /* 20ms */
+
+#define TSCHED_WATERMARK_INC_STEP_USEC (10*PA_USEC_PER_MSEC)       /* 10ms  */
+#define TSCHED_WATERMARK_DEC_STEP_USEC (5*PA_USEC_PER_MSEC)        /* 5ms */
+#define TSCHED_WATERMARK_VERIFY_AFTER_USEC (20*PA_USEC_PER_SEC)    /* 20s */
+#define TSCHED_WATERMARK_INC_THRESHOLD_USEC (1*PA_USEC_PER_MSEC)   /* 3ms */
+#define TSCHED_WATERMARK_DEC_THRESHOLD_USEC (100*PA_USEC_PER_MSEC) /* 100ms */
+#define TSCHED_WATERMARK_STEP_USEC (10*PA_USEC_PER_MSEC)           /* 10ms */
+
+#define TSCHED_MIN_SLEEP_USEC (10*PA_USEC_PER_MSEC)                /* 10ms */
+#define TSCHED_MIN_WAKEUP_USEC (4*PA_USEC_PER_MSEC)                /* 4ms */
+
+#define SMOOTHER_MIN_INTERVAL (2*PA_USEC_PER_MSEC)                 /* 2ms */
+#define SMOOTHER_MAX_INTERVAL (200*PA_USEC_PER_MSEC)               /* 200ms */
 
 #define VOLUME_ACCURACY (PA_VOLUME_NORM/100)
 
@@ -93,7 +104,12 @@ struct userdata {
         hwbuf_unused,
         min_sleep,
         min_wakeup,
-        watermark_step;
+        watermark_inc_step,
+        watermark_dec_step,
+        watermark_inc_threshold,
+        watermark_dec_threshold;
+
+    pa_usec_t watermark_dec_not_before;
 
     unsigned nfragments;
 
@@ -108,6 +124,8 @@ struct userdata {
 
     pa_smoother *smoother;
     uint64_t read_count;
+    pa_usec_t smoother_interval;
+    pa_usec_t last_smoother_update;
 
     pa_reserve_wrapper *reserve;
     pa_hook_slot *reserve_slot;
@@ -236,6 +254,7 @@ static int reserve_monitor_init(struct userdata *u, const char *dname) {
 static void fix_min_sleep_wakeup(struct userdata *u) {
     size_t max_use, max_use_2;
     pa_assert(u);
+    pa_assert(u->use_tsched);
 
     max_use = u->hwbuf_size - u->hwbuf_unused;
     max_use_2 = pa_frame_align(max_use/2, &u->source->sample_spec);
@@ -250,6 +269,7 @@ static void fix_min_sleep_wakeup(struct userdata *u) {
 static void fix_tsched_watermark(struct userdata *u) {
     size_t max_use;
     pa_assert(u);
+    pa_assert(u->use_tsched);
 
     max_use = u->hwbuf_size - u->hwbuf_unused;
 
@@ -260,7 +280,7 @@ static void fix_tsched_watermark(struct userdata *u) {
         u->tsched_watermark = u->min_wakeup;
 }
 
-static void adjust_after_overrun(struct userdata *u) {
+static void increase_watermark(struct userdata *u) {
     size_t old_watermark;
     pa_usec_t old_min_latency, new_min_latency;
 
@@ -269,36 +289,72 @@ static void adjust_after_overrun(struct userdata *u) {
 
     /* First, just try to increase the watermark */
     old_watermark = u->tsched_watermark;
-    u->tsched_watermark = PA_MIN(u->tsched_watermark * 2, u->tsched_watermark + u->watermark_step);
-
+    u->tsched_watermark = PA_MIN(u->tsched_watermark * 2, u->tsched_watermark + u->watermark_inc_step);
     fix_tsched_watermark(u);
 
     if (old_watermark != u->tsched_watermark) {
-        pa_log_notice("Increasing wakeup watermark to %0.2f ms",
-                      (double) pa_bytes_to_usec(u->tsched_watermark, &u->source->sample_spec) / PA_USEC_PER_MSEC);
+        pa_log_info("Increasing wakeup watermark to %0.2f ms",
+                    (double) pa_bytes_to_usec(u->tsched_watermark, &u->source->sample_spec) / PA_USEC_PER_MSEC);
         return;
     }
 
     /* Hmm, we cannot increase the watermark any further, hence let's raise the latency */
     old_min_latency = u->source->thread_info.min_latency;
-    new_min_latency = PA_MIN(old_min_latency * 2, old_min_latency + TSCHED_WATERMARK_STEP_USEC);
+    new_min_latency = PA_MIN(old_min_latency * 2, old_min_latency + TSCHED_WATERMARK_INC_STEP_USEC);
     new_min_latency = PA_MIN(new_min_latency, u->source->thread_info.max_latency);
 
     if (old_min_latency != new_min_latency) {
-        pa_log_notice("Increasing minimal latency to %0.2f ms",
-                      (double) new_min_latency / PA_USEC_PER_MSEC);
+        pa_log_info("Increasing minimal latency to %0.2f ms",
+                    (double) new_min_latency / PA_USEC_PER_MSEC);
 
         pa_source_set_latency_range_within_thread(u->source, new_min_latency, u->source->thread_info.max_latency);
-        return;
     }
 
     /* When we reach this we're officialy fucked! */
 }
 
+static void decrease_watermark(struct userdata *u) {
+    size_t old_watermark;
+    pa_usec_t now;
+
+    pa_assert(u);
+    pa_assert(u->use_tsched);
+
+    now = pa_rtclock_now();
+
+    if (u->watermark_dec_not_before <= 0)
+        goto restart;
+
+    if (u->watermark_dec_not_before > now)
+        return;
+
+    old_watermark = u->tsched_watermark;
+
+    if (u->tsched_watermark < u->watermark_dec_step)
+        u->tsched_watermark = u->tsched_watermark / 2;
+    else
+        u->tsched_watermark = PA_MAX(u->tsched_watermark / 2, u->tsched_watermark - u->watermark_dec_step);
+
+    fix_tsched_watermark(u);
+
+    if (old_watermark != u->tsched_watermark)
+        pa_log_info("Decreasing wakeup watermark to %0.2f ms",
+                    (double) pa_bytes_to_usec(u->tsched_watermark, &u->source->sample_spec) / PA_USEC_PER_MSEC);
+
+    /* We don't change the latency range*/
+
+restart:
+    u->watermark_dec_not_before = now + TSCHED_WATERMARK_VERIFY_AFTER_USEC;
+}
+
 static pa_usec_t hw_sleep_time(struct userdata *u, pa_usec_t *sleep_usec, pa_usec_t*process_usec) {
     pa_usec_t wm, usec;
 
+    pa_assert(sleep_usec);
+    pa_assert(process_usec);
+
     pa_assert(u);
+    pa_assert(u->use_tsched);
 
     usec = pa_source_get_requested_latency_within_thread(u->source);
 
@@ -347,7 +403,7 @@ static int try_recover(struct userdata *u, const char *call, int err) {
     return 0;
 }
 
-static size_t check_left_to_record(struct userdata *u, size_t n_bytes) {
+static size_t check_left_to_record(struct userdata *u, size_t n_bytes, pa_bool_t on_timeout) {
     size_t left_to_record;
     size_t rec_space = u->hwbuf_size - u->hwbuf_unused;
 
@@ -356,14 +412,11 @@ static size_t check_left_to_record(struct userdata *u, size_t n_bytes) {
      * it is removed from the buffer. This is particularly important
      * when block transfer is used. */
 
-    if (n_bytes <= rec_space) {
+    if (n_bytes <= rec_space)
         left_to_record = rec_space - n_bytes;
+    else {
 
-#ifdef DEBUG_TIMING
-        pa_log_debug("%0.2f ms left to record", (double) pa_bytes_to_usec(left_to_record, &u->source->sample_spec) / PA_USEC_PER_MSEC);
-#endif
-
-    } else {
+        /* We got a dropout. What a mess! */
         left_to_record = 0;
 
 #ifdef DEBUG_TIMING
@@ -372,15 +425,36 @@ static size_t check_left_to_record(struct userdata *u, size_t n_bytes) {
 
         if (pa_log_ratelimit())
             pa_log_info("Overrun!");
+    }
 
-        if (u->use_tsched)
-            adjust_after_overrun(u);
+#ifdef DEBUG_TIMING
+    pa_log_debug("%0.2f ms left to record", (double) pa_bytes_to_usec(left_to_record, &u->source->sample_spec) / PA_USEC_PER_MSEC);
+#endif
+
+    if (u->use_tsched) {
+        pa_bool_t reset_not_before = TRUE;
+
+        if (left_to_record < u->watermark_inc_threshold)
+            increase_watermark(u);
+        else if (left_to_record > u->watermark_dec_threshold) {
+            reset_not_before = FALSE;
+
+            /* We decrease the watermark only if have actually been
+             * woken up by a timeout. If something else woke us up
+             * it's too easy to fulfill the deadlines... */
+
+            if (on_timeout)
+                decrease_watermark(u);
+        }
+
+        if (reset_not_before)
+            u->watermark_dec_not_before = 0;
     }
 
     return left_to_record;
 }
 
-static int mmap_read(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled) {
+static int mmap_read(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled, pa_bool_t on_timeout) {
     pa_bool_t work_done = FALSE;
     pa_usec_t max_sleep_usec = 0, process_usec = 0;
     size_t left_to_record;
@@ -412,7 +486,8 @@ static int mmap_read(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled
         pa_log_debug("avail: %lu", (unsigned long) n_bytes);
 #endif
 
-        left_to_record = check_left_to_record(u, n_bytes);
+        left_to_record = check_left_to_record(u, n_bytes, on_timeout);
+        on_timeout = FALSE;
 
         if (u->use_tsched)
             if (!polled &&
@@ -538,7 +613,7 @@ static int mmap_read(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled
     return work_done ? 1 : 0;
 }
 
-static int unix_read(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled) {
+static int unix_read(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled, pa_bool_t on_timeout) {
     int work_done = FALSE;
     pa_usec_t max_sleep_usec = 0, process_usec = 0;
     size_t left_to_record;
@@ -565,7 +640,8 @@ static int unix_read(struct userdata *u, pa_usec_t *sleep_usec, pa_bool_t polled
         }
 
         n_bytes = (size_t) n * u->frame_size;
-        left_to_record = check_left_to_record(u, n_bytes);
+        left_to_record = check_left_to_record(u, n_bytes, on_timeout);
+        on_timeout = FALSE;
 
         if (u->use_tsched)
             if (!polled &&
@@ -691,15 +767,23 @@ static void update_smoother(struct userdata *u) {
         now1 = pa_timespec_load(&htstamp);
     }
 
-    position = u->read_count + ((uint64_t) delay * (uint64_t) u->frame_size);
-
     /* Hmm, if the timestamp is 0, then it wasn't set and we take the current time */
     if (now1 <= 0)
         now1 = pa_rtclock_now();
 
+    /* check if the time since the last update is bigger than the interval */
+    if (u->last_smoother_update > 0)
+        if (u->last_smoother_update + u->smoother_interval > now1)
+            return;
+
+    position = u->read_count + ((uint64_t) delay * (uint64_t) u->frame_size);
     now2 = pa_bytes_to_usec(position, &u->source->sample_spec);
 
     pa_smoother_put(u->smoother, now1, now2);
+
+    u->last_smoother_update = now1;
+    /* exponentially increase the update interval up to the MAX limit */
+    u->smoother_interval = PA_MIN (u->smoother_interval * 2, SMOOTHER_MAX_INTERVAL);
 }
 
 static pa_usec_t source_get_latency(struct userdata *u) {
@@ -862,6 +946,8 @@ static int unsuspend(struct userdata *u) {
 
     u->read_count = 0;
     pa_smoother_reset(u->smoother, pa_rtclock_now(), TRUE);
+    u->smoother_interval = SMOOTHER_MIN_INTERVAL;
+    u->last_smoother_update = 0;
 
     pa_log_info("Resumed successfully...");
 
@@ -1143,11 +1229,12 @@ static void thread_func(void *userdata) {
         if (PA_SOURCE_IS_OPENED(u->source->thread_info.state)) {
             int work_done;
             pa_usec_t sleep_usec = 0;
+            pa_bool_t on_timeout = pa_rtpoll_timer_elapsed(u->rtpoll);
 
             if (u->use_mmap)
-                work_done = mmap_read(u, &sleep_usec, revents & POLLIN);
+                work_done = mmap_read(u, &sleep_usec, revents & POLLIN, on_timeout);
             else
-                work_done = unix_read(u, &sleep_usec, revents & POLLIN);
+                work_done = unix_read(u, &sleep_usec, revents & POLLIN, on_timeout);
 
             if (work_done < 0)
                 goto fail;
@@ -1469,6 +1556,7 @@ pa_source *pa_alsa_source_new(pa_module *m, pa_modargs *ma, const char*driver, p
             5,
             pa_rtclock_now(),
             FALSE);
+    u->smoother_interval = SMOOTHER_MIN_INTERVAL;
 
     dev_id = pa_modargs_get_value(
             ma, "device_id",
@@ -1616,7 +1704,6 @@ pa_source *pa_alsa_source_new(pa_module *m, pa_modargs *ma, const char*driver, p
     u->fragment_size = frag_size = (uint32_t) (period_frames * frame_size);
     u->nfragments = nfrags;
     u->hwbuf_size = u->fragment_size * nfrags;
-    u->tsched_watermark = pa_usec_to_bytes_round_up(pa_bytes_to_usec_round_up(tsched_watermark, &requested_ss), &u->source->sample_spec);
     pa_cvolume_mute(&u->hardware_volume, u->source->sample_spec.channels);
 
     pa_log_info("Using %u fragments of size %lu bytes, buffer time is %0.2fms",
@@ -1624,7 +1711,13 @@ pa_source *pa_alsa_source_new(pa_module *m, pa_modargs *ma, const char*driver, p
                 (double) pa_bytes_to_usec(u->hwbuf_size, &ss) / PA_USEC_PER_MSEC);
 
     if (u->use_tsched) {
-        u->watermark_step = pa_usec_to_bytes(TSCHED_WATERMARK_STEP_USEC, &u->source->sample_spec);
+        u->tsched_watermark = pa_usec_to_bytes_round_up(pa_bytes_to_usec_round_up(tsched_watermark, &requested_ss), &u->source->sample_spec);
+
+        u->watermark_inc_step = pa_usec_to_bytes(TSCHED_WATERMARK_INC_STEP_USEC, &u->source->sample_spec);
+        u->watermark_dec_step = pa_usec_to_bytes(TSCHED_WATERMARK_DEC_STEP_USEC, &u->source->sample_spec);
+
+        u->watermark_inc_threshold = pa_usec_to_bytes_round_up(TSCHED_WATERMARK_INC_THRESHOLD_USEC, &u->source->sample_spec);
+        u->watermark_dec_threshold = pa_usec_to_bytes_round_up(TSCHED_WATERMARK_DEC_THRESHOLD_USEC, &u->source->sample_spec);
 
         fix_min_sleep_wakeup(u);
         fix_tsched_watermark(u);
diff --git a/src/modules/alsa/mixer/paths/analog-output-lfe-on-mono.conf b/src/modules/alsa/mixer/paths/analog-output-lfe-on-mono.conf
index 2db976a5..3457d4a2 100644
--- a/src/modules/alsa/mixer/paths/analog-output-lfe-on-mono.conf
+++ b/src/modules/alsa/mixer/paths/analog-output-lfe-on-mono.conf
@@ -41,9 +41,12 @@ volume = merge
 override-map.1 = lfe
 override-map.2 = lfe,lfe
 
+; This profile path is intended to control the speaker, not the
+; headphones. But it should not hurt if we leave the headphone jack
+; enabled nonetheless.
 [Element Headphone]
-switch = off
-volume = off
+switch = mute
+volume = zero
 
 [Element Speaker]
 switch = mute
diff --git a/src/modules/alsa/mixer/paths/analog-output-mono.conf b/src/modules/alsa/mixer/paths/analog-output-mono.conf
index a58cc970..dc270cfe 100644
--- a/src/modules/alsa/mixer/paths/analog-output-mono.conf
+++ b/src/modules/alsa/mixer/paths/analog-output-mono.conf
@@ -38,9 +38,12 @@ volume = merge
 override-map.1 = all
 override-map.2 = all-left,all-right
 
+; This profile path is intended to control the speaker, not the
+; headphones. But it should not hurt if we leave the headphone jack
+; enabled nonetheless.
 [Element Headphone]
-switch = off
-volume = off
+switch = mute
+volume = zero
 
 [Element Speaker]
 switch = mute
diff --git a/src/modules/alsa/mixer/paths/analog-output.conf b/src/modules/alsa/mixer/paths/analog-output.conf
index b412a437..f71a05a1 100644
--- a/src/modules/alsa/mixer/paths/analog-output.conf
+++ b/src/modules/alsa/mixer/paths/analog-output.conf
@@ -37,9 +37,12 @@ override-map.2 = all-left,all-right
 switch = off
 volume = off
 
+; This profile path is intended to control the speaker, not the
+; headphones. But it should not hurt if we leave the headphone jack
+; enabled nonetheless.
 [Element Headphone]
-switch = off
-volume = off
+switch = mute
+volume = zero
 
 [Element Speaker]
 switch = mute
diff --git a/src/modules/gconf/module-gconf.c b/src/modules/gconf/module-gconf.c
index c01ebbf6..85523b39 100644
--- a/src/modules/gconf/module-gconf.c
+++ b/src/modules/gconf/module-gconf.c
@@ -52,9 +52,6 @@ PA_MODULE_LOAD_ONCE(TRUE);
 #define MAX_MODULES 10
 #define BUF_MAX 2048
 
-/* #undef PA_GCONF_HELPER */
-/* #define PA_GCONF_HELPER "/home/lennart/projects/pulseaudio/src/gconf-helper" */
-
 struct module_item {
     char *name;
     char *args;
@@ -343,7 +340,11 @@ int pa__init(pa_module*m) {
     u->io_event = NULL;
     u->buf_fill = 0;
 
-    if ((u->fd = pa_start_child_for_read(PA_GCONF_HELPER, NULL, &u->pid)) < 0)
+    if ((u->fd = pa_start_child_for_read(
+#if defined(__linux__) && !defined(__OPTIMIZE__)
+                              pa_run_from_build_tree() ? PA_BUILDDIR "/.libs/gconf-helper" :
+#endif
+                 PA_GCONF_HELPER, NULL, &u->pid)) < 0)
         goto fail;
 
     u->io_event = m->core->mainloop->io_new(
diff --git a/src/modules/module-combine.c b/src/modules/module-combine.c
index 582cbce1..e90ef11c 100644
--- a/src/modules/module-combine.c
+++ b/src/modules/module-combine.c
@@ -1161,6 +1161,8 @@ int pa__init(pa_module*m) {
         pa_channel_map slaves_map;
         pa_bool_t is_first_slave = TRUE;
 
+        pa_sample_spec_init(&slaves_spec);
+
         while ((n = pa_split(slaves, ",", &split_state))) {
             pa_sink *slave_sink;
 
diff --git a/src/modules/module-ladspa-sink.c b/src/modules/module-ladspa-sink.c
index f2d53d00..aa28f7fd 100644
--- a/src/modules/module-ladspa-sink.c
+++ b/src/modules/module-ladspa-sink.c
@@ -99,7 +99,7 @@ static const char* const valid_modargs[] = {
 };
 
 /* Called from I/O thread context */
-static int sink_process_msg(pa_msgobject *o, int code, void *data, int64_t offset, pa_memchunk *chunk) {
+static int sink_process_msg_cb(pa_msgobject *o, int code, void *data, int64_t offset, pa_memchunk *chunk) {
     struct userdata *u = PA_SINK(o)->userdata;
 
     switch (code) {
@@ -130,7 +130,7 @@ static int sink_process_msg(pa_msgobject *o, int code, void *data, int64_t offse
 }
 
 /* Called from main context */
-static int sink_set_state(pa_sink *s, pa_sink_state_t state) {
+static int sink_set_state_cb(pa_sink *s, pa_sink_state_t state) {
     struct userdata *u;
 
     pa_sink_assert_ref(s);
@@ -145,7 +145,7 @@ static int sink_set_state(pa_sink *s, pa_sink_state_t state) {
 }
 
 /* Called from I/O thread context */
-static void sink_request_rewind(pa_sink *s) {
+static void sink_request_rewind_cb(pa_sink *s) {
     struct userdata *u;
 
     pa_sink_assert_ref(s);
@@ -160,7 +160,7 @@ static void sink_request_rewind(pa_sink *s) {
 }
 
 /* Called from I/O thread context */
-static void sink_update_requested_latency(pa_sink *s) {
+static void sink_update_requested_latency_cb(pa_sink *s) {
     struct userdata *u;
 
     pa_sink_assert_ref(s);
@@ -176,6 +176,34 @@ static void sink_update_requested_latency(pa_sink *s) {
             pa_sink_get_requested_latency_within_thread(s));
 }
 
+/* Called from main context */
+static void sink_set_volume_cb(pa_sink *s) {
+    struct userdata *u;
+
+    pa_sink_assert_ref(s);
+    pa_assert_se(u = s->userdata);
+
+    if (!PA_SINK_IS_LINKED(pa_sink_get_state(s)) ||
+        !PA_SINK_INPUT_IS_LINKED(pa_sink_input_get_state(u->sink_input)))
+        return;
+
+    pa_sink_input_set_volume(u->sink_input, &s->real_volume, s->save_volume, TRUE);
+}
+
+/* Called from main context */
+static void sink_set_mute_cb(pa_sink *s) {
+    struct userdata *u;
+
+    pa_sink_assert_ref(s);
+    pa_assert_se(u = s->userdata);
+
+    if (!PA_SINK_IS_LINKED(pa_sink_get_state(s)) ||
+        !PA_SINK_INPUT_IS_LINKED(pa_sink_input_get_state(u->sink_input)))
+        return;
+
+    pa_sink_input_set_mute(u->sink_input, s->muted, s->save_muted);
+}
+
 /* Called from I/O thread context */
 static int sink_input_pop_cb(pa_sink_input *i, size_t nbytes, pa_memchunk *chunk) {
     struct userdata *u;
@@ -390,8 +418,31 @@ static void sink_input_moving_cb(pa_sink_input *i, pa_sink *dest) {
     pa_sink_input_assert_ref(i);
     pa_assert_se(u = i->userdata);
 
-    pa_sink_set_asyncmsgq(u->sink, dest->asyncmsgq);
-    pa_sink_update_flags(u->sink, PA_SINK_LATENCY|PA_SINK_DYNAMIC_LATENCY, dest->flags);
+    if (dest) {
+        pa_sink_set_asyncmsgq(u->sink, dest->asyncmsgq);
+        pa_sink_update_flags(u->sink, PA_SINK_LATENCY|PA_SINK_DYNAMIC_LATENCY, dest->flags);
+    } else
+        pa_sink_set_asyncmsgq(u->sink, NULL);
+}
+
+/* Called from main context */
+static void sink_input_volume_changed_cb(pa_sink_input *i) {
+    struct userdata *u;
+
+    pa_sink_input_assert_ref(i);
+    pa_assert_se(u = i->userdata);
+
+    pa_sink_volume_changed(u->sink, &i->volume);
+}
+
+/* Called from main context */
+static void sink_input_mute_changed_cb(pa_sink_input *i) {
+    struct userdata *u;
+
+    pa_sink_input_assert_ref(i);
+    pa_assert_se(u = i->userdata);
+
+    pa_sink_mute_changed(u->sink, i->muted);
 }
 
 int pa__init(pa_module*m) {
@@ -731,7 +782,9 @@ int pa__init(pa_module*m) {
         goto fail;
     }
 
-    u->sink = pa_sink_new(m->core, &sink_data, master->flags & (PA_SINK_LATENCY|PA_SINK_DYNAMIC_LATENCY));
+    u->sink = pa_sink_new(m->core, &sink_data,
+                          PA_SINK_HW_MUTE_CTRL|PA_SINK_HW_VOLUME_CTRL|PA_SINK_DECIBEL_VOLUME|
+                          (master->flags & (PA_SINK_LATENCY|PA_SINK_DYNAMIC_LATENCY)));
     pa_sink_new_data_done(&sink_data);
 
     if (!u->sink) {
@@ -739,10 +792,12 @@ int pa__init(pa_module*m) {
         goto fail;
     }
 
-    u->sink->parent.process_msg = sink_process_msg;
-    u->sink->set_state = sink_set_state;
-    u->sink->update_requested_latency = sink_update_requested_latency;
-    u->sink->request_rewind = sink_request_rewind;
+    u->sink->parent.process_msg = sink_process_msg_cb;
+    u->sink->set_state = sink_set_state_cb;
+    u->sink->update_requested_latency = sink_update_requested_latency_cb;
+    u->sink->request_rewind = sink_request_rewind_cb;
+    u->sink->set_volume = sink_set_volume_cb;
+    u->sink->set_mute = sink_set_mute_cb;
     u->sink->userdata = u;
 
     pa_sink_set_asyncmsgq(u->sink, master->asyncmsgq);
@@ -775,6 +830,8 @@ int pa__init(pa_module*m) {
     u->sink_input->state_change = sink_input_state_change_cb;
     u->sink_input->may_move_to = sink_input_may_move_to_cb;
     u->sink_input->moving = sink_input_moving_cb;
+    u->sink_input->volume_changed = sink_input_volume_changed_cb;
+    u->sink_input->mute_changed = sink_input_mute_changed_cb;
     u->sink_input->userdata = u;
 
     pa_sink_put(u->sink);
diff --git a/src/modules/module-remap-sink.c b/src/modules/module-remap-sink.c
index 0b4fdc9b..becff55c 100644
--- a/src/modules/module-remap-sink.c
+++ b/src/modules/module-remap-sink.c
@@ -302,8 +302,11 @@ static void sink_input_moving_cb(pa_sink_input *i, pa_sink *dest) {
     pa_sink_input_assert_ref(i);
     pa_assert_se(u = i->userdata);
 
-    pa_sink_set_asyncmsgq(u->sink, dest->asyncmsgq);
-    pa_sink_update_flags(u->sink, PA_SINK_LATENCY|PA_SINK_DYNAMIC_LATENCY, dest->flags);
+    if (dest) {
+        pa_sink_set_asyncmsgq(u->sink, dest->asyncmsgq);
+        pa_sink_update_flags(u->sink, PA_SINK_LATENCY|PA_SINK_DYNAMIC_LATENCY, dest->flags);
+    } else
+        pa_sink_set_asyncmsgq(u->sink, NULL);
 }
 
 int pa__init(pa_module*m) {
diff --git a/src/modules/module-solaris.c b/src/modules/module-solaris.c
index 0920d25e..71f14071 100644
--- a/src/modules/module-solaris.c
+++ b/src/modules/module-solaris.c
@@ -60,6 +60,7 @@
 #include <pulsecore/thread-mq.h>
 #include <pulsecore/rtpoll.h>
 #include <pulsecore/thread.h>
+#include <pulsecore/time-smoother.h>
 
 #include "module-solaris-symdef.h"
 
@@ -110,6 +111,8 @@ struct userdata {
     uint32_t prev_playback_samples, prev_record_samples;
 
     int32_t minimum_request;
+
+    pa_smoother *smoother;
 };
 
 static const char* const valid_modargs[] = {
@@ -133,6 +136,9 @@ static const char* const valid_modargs[] = {
 #define MAX_RENDER_HZ   (300)
 /* This render rate limit imposes a minimum latency, but without it we waste too much CPU time. */
 
+#define MAX_BUFFER_SIZE (128 * 1024)
+/* An attempt to buffer more than 128 KB causes write() to fail with errno == EAGAIN. */
+
 static uint64_t get_playback_buffered_bytes(struct userdata *u) {
     audio_info_t info;
     uint64_t played_bytes;
@@ -145,7 +151,12 @@ static uint64_t get_playback_buffered_bytes(struct userdata *u) {
 
     /* Handle wrap-around of the device's sample counter, which is a uint_32. */
     if (u->prev_playback_samples > info.play.samples) {
-        /* Unfortunately info.play.samples can sometimes go backwards, even before it wraps! */
+        /*
+         * Unfortunately info.play.samples can sometimes go backwards, even before it wraps!
+         * The bug seems to be absent on Solaris x86 nv117 with audio810 driver, at least on this (UP) machine.
+         * The bug is present on a different (SMP) machine running Solaris x86 nv103 with audioens driver.
+         * An earlier revision of this file mentions the same bug independently (unknown configuration).
+         */
         if (u->prev_playback_samples + info.play.samples < 240000) {
             ++u->play_samples_msw;
         } else {
@@ -155,6 +166,8 @@ static uint64_t get_playback_buffered_bytes(struct userdata *u) {
     u->prev_playback_samples = info.play.samples;
     played_bytes = (((uint64_t)u->play_samples_msw << 32) + info.play.samples) * u->frame_size;
 
+    pa_smoother_put(u->smoother, pa_rtclock_now(), pa_bytes_to_usec(played_bytes, &u->sink->sample_spec));
+
     return u->written_bytes - played_bytes;
 }
 
@@ -387,6 +400,8 @@ static int sink_process_msg(pa_msgobject *o, int code, void *data, int64_t offse
 
                     pa_assert(PA_SINK_IS_OPENED(u->sink->thread_info.state));
 
+                    pa_smoother_pause(u->smoother, pa_rtclock_now());
+
                     if (!u->source || u->source_suspended) {
                         if (suspend(u) < 0)
                             return -1;
@@ -398,6 +413,8 @@ static int sink_process_msg(pa_msgobject *o, int code, void *data, int64_t offse
                 case PA_SINK_RUNNING:
 
                     if (u->sink->thread_info.state == PA_SINK_SUSPENDED) {
+                        pa_smoother_resume(u->smoother, pa_rtclock_now(), TRUE);
+
                         if (!u->source || u->source_suspended) {
                             if (unsuspend(u) < 0)
                                 return -1;
@@ -479,7 +496,7 @@ static void sink_set_volume(pa_sink *s) {
     if (u->fd >= 0) {
         AUDIO_INITINFO(&info);
 
-        info.play.gain = pa_cvolume_max(&s->virtual_volume) * AUDIO_MAX_GAIN / PA_VOLUME_NORM;
+        info.play.gain = pa_cvolume_max(&s->real_volume) * AUDIO_MAX_GAIN / PA_VOLUME_NORM;
         assert(info.play.gain <= AUDIO_MAX_GAIN);
 
         if (ioctl(u->fd, AUDIO_SETINFO, &info) < 0) {
@@ -501,8 +518,7 @@ static void sink_get_volume(pa_sink *s) {
         if (ioctl(u->fd, AUDIO_GETINFO, &info) < 0)
             pa_log("AUDIO_SETINFO: %s", pa_cstrerror(errno));
         else
-            pa_cvolume_set(&s->virtual_volume, s->sample_spec.channels,
-                info.play.gain * PA_VOLUME_NORM / AUDIO_MAX_GAIN);
+            pa_cvolume_set(&s->real_volume, s->sample_spec.channels, info.play.gain * PA_VOLUME_NORM / AUDIO_MAX_GAIN);
     }
 }
 
@@ -515,7 +531,7 @@ static void source_set_volume(pa_source *s) {
     if (u->fd >= 0) {
         AUDIO_INITINFO(&info);
 
-        info.play.gain = pa_cvolume_max(&s->virtual_volume) * AUDIO_MAX_GAIN / PA_VOLUME_NORM;
+        info.play.gain = pa_cvolume_max(&s->volume) * AUDIO_MAX_GAIN / PA_VOLUME_NORM;
         assert(info.play.gain <= AUDIO_MAX_GAIN);
 
         if (ioctl(u->fd, AUDIO_SETINFO, &info) < 0) {
@@ -537,8 +553,7 @@ static void source_get_volume(pa_source *s) {
         if (ioctl(u->fd, AUDIO_GETINFO, &info) < 0)
             pa_log("AUDIO_SETINFO: %s", pa_cstrerror(errno));
         else
-            pa_cvolume_set(&s->virtual_volume, s->sample_spec.channels,
-                info.play.gain * PA_VOLUME_NORM / AUDIO_MAX_GAIN);
+            pa_cvolume_set(&s->volume, s->sample_spec.channels, info.play.gain * PA_VOLUME_NORM / AUDIO_MAX_GAIN);
     }
 }
 
@@ -606,11 +621,13 @@ static void thread_func(void *userdata) {
 
     pa_thread_mq_install(&u->thread_mq);
 
+    pa_smoother_set_time_offset(u->smoother, pa_rtclock_now());
+
     for (;;) {
         /* Render some data and write it to the dsp */
 
         if (u->sink && PA_SINK_IS_OPENED(u->sink->thread_info.state)) {
-            pa_usec_t xtime0;
+            pa_usec_t xtime0, ysleep_interval, xsleep_interval;
             uint64_t buffered_bytes;
 
             if (u->sink->thread_info.rewind_requested)
@@ -629,12 +646,15 @@ static void thread_func(void *userdata) {
                 info.play.error = 0;
                 if (ioctl(u->fd, AUDIO_SETINFO, &info) < 0)
                     pa_log("AUDIO_SETINFO: %s", pa_cstrerror(errno));
+
+                pa_smoother_reset(u->smoother, pa_rtclock_now(), TRUE);
             }
 
             for (;;) {
                 void *p;
                 ssize_t w;
                 size_t len;
+                int write_type = 1;
 
                 /*
                  * Since we cannot modify the size of the output buffer we fake it
@@ -652,38 +672,31 @@ static void thread_func(void *userdata) {
                     break;
 
                 if (u->memchunk.length < len)
-                    pa_sink_render(u->sink, u->sink->thread_info.max_request, &u->memchunk);
+                    pa_sink_render(u->sink, len - u->memchunk.length, &u->memchunk);
+
+                len = PA_MIN(u->memchunk.length, len);
 
                 p = pa_memblock_acquire(u->memchunk.memblock);
-                w = pa_write(u->fd, (uint8_t*) p + u->memchunk.index, u->memchunk.length, NULL);
+                w = pa_write(u->fd, (uint8_t*) p + u->memchunk.index, len, &write_type);
                 pa_memblock_release(u->memchunk.memblock);
 
                 if (w <= 0) {
-                    switch (errno) {
-                        case EINTR:
-                            continue;
-                        case EAGAIN:
-                            /* If the buffer_size is too big, we get EAGAIN. Avoiding that limit by trial and error
-                             * is not ideal, but I don't know how to get the system to tell me what the limit is.
-                             */
-                            u->buffer_size = u->buffer_size * 18 / 25;
-                            u->buffer_size -= u->buffer_size % u->frame_size;
-                            u->buffer_size = PA_MAX(u->buffer_size, 2 * u->minimum_request);
-                            pa_sink_set_max_request_within_thread(u->sink, u->buffer_size);
-                            pa_sink_set_max_rewind_within_thread(u->sink, u->buffer_size);
-                            pa_log("EAGAIN. Buffer size is now %u bytes (%llu buffered)", u->buffer_size, buffered_bytes);
-                            break;
-                        default:
-                            pa_log("Failed to write data to DSP: %s", pa_cstrerror(errno));
-                            goto fail;
+                    if (errno == EINTR) {
+                        continue;
+                    } else if (errno == EAGAIN) {
+                        /* We may have realtime priority so yield the CPU to ensure that fd can become writable again. */
+                        pa_log_debug("EAGAIN with %llu bytes buffered.", buffered_bytes);
+                        break;
+                    } else {
+                        pa_log("Failed to write data to DSP: %s", pa_cstrerror(errno));
+                        goto fail;
                     }
                 } else {
                     pa_assert(w % u->frame_size == 0);
 
                     u->written_bytes += w;
-                    u->memchunk.length -= w;
-
                     u->memchunk.index += w;
+                    u->memchunk.length -= w;
                     if (u->memchunk.length <= 0) {
                         pa_memblock_unref(u->memchunk.memblock);
                         pa_memchunk_reset(&u->memchunk);
@@ -691,7 +704,9 @@ static void thread_func(void *userdata) {
                 }
             }
 
-            pa_rtpoll_set_timer_absolute(u->rtpoll, xtime0 + pa_bytes_to_usec(buffered_bytes / 2, &u->sink->sample_spec));
+            ysleep_interval = pa_bytes_to_usec(buffered_bytes / 2, &u->sink->sample_spec);
+            xsleep_interval = pa_smoother_translate(u->smoother, xtime0, ysleep_interval);
+            pa_rtpoll_set_timer_absolute(u->rtpoll, xtime0 + PA_MIN(xsleep_interval, ysleep_interval));
         } else
             pa_rtpoll_set_timer_disabled(u->rtpoll);
 
@@ -797,7 +812,7 @@ static void sig_callback(pa_mainloop_api *api, pa_signal_event*e, int sig, void
     pa_log_debug("caught signal");
 
     if (u->sink) {
-        pa_sink_get_volume(u->sink, TRUE, FALSE);
+        pa_sink_get_volume(u->sink, TRUE);
         pa_sink_get_mute(u->sink, TRUE);
     }
 
@@ -812,7 +827,7 @@ int pa__init(pa_module *m) {
     pa_channel_map map;
     pa_modargs *ma = NULL;
     uint32_t buffer_length_msec;
-    int fd;
+    int fd = -1;
     pa_sink_new_data sink_new_data;
     pa_source_new_data source_new_data;
     char const *name;
@@ -838,6 +853,9 @@ int pa__init(pa_module *m) {
 
     u = pa_xnew0(struct userdata, 1);
 
+    if (!(u->smoother = pa_smoother_new(PA_USEC_PER_SEC, PA_USEC_PER_SEC * 2, TRUE, TRUE, 10, pa_rtclock_now(), TRUE)))
+        goto fail;
+
     /*
      * For a process (or several processes) to use the same audio device for both
      * record and playback at the same time, the device's mixer must be enabled.
@@ -861,7 +879,13 @@ int pa__init(pa_module *m) {
     }
     u->buffer_size = pa_usec_to_bytes(1000 * buffer_length_msec, &ss);
     if (u->buffer_size < 2 * u->minimum_request) {
-        pa_log("supplied buffer size argument is too small");
+        pa_log("buffer_length argument cannot be smaller than %u",
+               (unsigned)(pa_bytes_to_usec(2 * u->minimum_request, &ss) / 1000));
+        goto fail;
+    }
+    if (u->buffer_size > MAX_BUFFER_SIZE) {
+        pa_log("buffer_length argument cannot be greater than %u",
+               (unsigned)(pa_bytes_to_usec(MAX_BUFFER_SIZE, &ss) / 1000));
         goto fail;
     }
 
@@ -924,6 +948,7 @@ int pa__init(pa_module *m) {
 
         pa_source_set_asyncmsgq(u->source, u->thread_mq.inq);
         pa_source_set_rtpoll(u->source, u->rtpoll);
+        pa_source_set_fixed_latency(u->source, pa_bytes_to_usec(u->buffer_size, &u->source->sample_spec));
 
         u->source->get_volume = source_get_volume;
         u->source->set_volume = source_set_volume;
@@ -966,15 +991,15 @@ int pa__init(pa_module *m) {
 
         pa_sink_set_asyncmsgq(u->sink, u->thread_mq.inq);
         pa_sink_set_rtpoll(u->sink, u->rtpoll);
+        pa_sink_set_fixed_latency(u->sink, pa_bytes_to_usec(u->buffer_size, &u->sink->sample_spec));
+        pa_sink_set_max_request(u->sink, u->buffer_size);
+        pa_sink_set_max_rewind(u->sink, u->buffer_size);
 
         u->sink->get_volume = sink_get_volume;
         u->sink->set_volume = sink_set_volume;
         u->sink->get_mute = sink_get_mute;
         u->sink->set_mute = sink_set_mute;
         u->sink->refresh_volume = u->sink->refresh_muted = TRUE;
-
-        pa_sink_set_max_request(u->sink, u->buffer_size);
-        pa_sink_set_max_rewind(u->sink, u->buffer_size);
     } else
         u->sink = NULL;
 
@@ -1075,6 +1100,9 @@ void pa__done(pa_module *m) {
     if (u->fd >= 0)
         close(u->fd);
 
+    if (u->smoother)
+        pa_smoother_free(u->smoother);
+
     pa_xfree(u->device_name);
 
     pa_xfree(u);
diff --git a/src/modules/module-udev-detect.c b/src/modules/module-udev-detect.c
index 0b30fd54..b41b9c0f 100644
--- a/src/modules/module-udev-detect.c
+++ b/src/modules/module-udev-detect.c
@@ -25,6 +25,7 @@
 
 #include <errno.h>
 #include <limits.h>
+#include <dirent.h>
 #include <sys/inotify.h>
 #include <libudev.h>
 
@@ -45,8 +46,9 @@ PA_MODULE_USAGE(
 
 struct device {
     char *path;
-    pa_bool_t accessible;
+    pa_bool_t need_verify;
     char *card_name;
+    char *args;
     uint32_t module;
 };
 
@@ -78,6 +80,7 @@ static void device_free(struct device *d) {
 
     pa_xfree(d->path);
     pa_xfree(d->card_name);
+    pa_xfree(d->args);
     pa_xfree(d);
 }
 
@@ -96,30 +99,166 @@ static const char *path_get_card_id(const char *path) {
     return e + 5;
 }
 
+static pa_bool_t is_card_busy(const char *id) {
+    char *card_path = NULL, *pcm_path = NULL, *sub_status = NULL;
+    DIR *card_dir = NULL, *pcm_dir = NULL;
+    FILE *status_file = NULL;
+    size_t len;
+    struct dirent *space = NULL, *de;
+    pa_bool_t busy = FALSE;
+    int r;
+
+    pa_assert(id);
+
+    card_path = pa_sprintf_malloc("/proc/asound/card%s", id);
+
+    if (!(card_dir = opendir(card_path))) {
+        pa_log_warn("Failed to open %s: %s", card_path, pa_cstrerror(errno));
+        goto fail;
+    }
+
+    len = offsetof(struct dirent, d_name) + fpathconf(dirfd(card_dir), _PC_NAME_MAX) + 1;
+    space = pa_xmalloc(len);
+
+    for (;;) {
+        de = NULL;
+
+        if ((r = readdir_r(card_dir, space, &de)) != 0) {
+            pa_log_warn("readdir_r() failed: %s", pa_cstrerror(r));
+            goto fail;
+        }
+
+        if (!de)
+            break;
+
+        if (!pa_startswith(de->d_name, "pcm"))
+            continue;
+
+        pa_xfree(pcm_path);
+        pcm_path = pa_sprintf_malloc("%s/%s", card_path, de->d_name);
+
+        if (pcm_dir)
+            closedir(pcm_dir);
+
+        if (!(pcm_dir = opendir(pcm_path))) {
+            pa_log_warn("Failed to open %s: %s", pcm_path, pa_cstrerror(errno));
+            continue;
+        }
+
+        for (;;) {
+            char line[32];
+
+            if ((r = readdir_r(pcm_dir, space, &de)) != 0) {
+                pa_log_warn("readdir_r() failed: %s", pa_cstrerror(r));
+                goto fail;
+            }
+
+            if (!de)
+                break;
+
+            if (!pa_startswith(de->d_name, "sub"))
+                continue;
+
+            pa_xfree(sub_status);
+            sub_status = pa_sprintf_malloc("%s/%s/status", pcm_path, de->d_name);
+
+            if (status_file)
+                fclose(status_file);
+
+            if (!(status_file = fopen(sub_status, "r"))) {
+                pa_log_warn("Failed to open %s: %s", sub_status, pa_cstrerror(errno));
+                continue;
+            }
+
+            if (!(fgets(line, sizeof(line)-1, status_file))) {
+                pa_log_warn("Failed to read from %s: %s", sub_status, pa_cstrerror(errno));
+                continue;
+            }
+
+            if (!pa_streq(line, "closed\n")) {
+                busy = TRUE;
+                break;
+            }
+        }
+    }
+
+fail:
+
+    pa_xfree(card_path);
+    pa_xfree(pcm_path);
+    pa_xfree(sub_status);
+    pa_xfree(space);
+
+    if (card_dir)
+        closedir(card_dir);
+
+    if (pcm_dir)
+        closedir(pcm_dir);
+
+    if (status_file)
+        fclose(status_file);
+
+    return busy;
+}
+
 static void verify_access(struct userdata *u, struct device *d) {
     char *cd;
     pa_card *card;
+    pa_bool_t accessible;
 
     pa_assert(u);
     pa_assert(d);
 
-    if (!(card = pa_namereg_get(u->core, d->card_name, PA_NAMEREG_CARD)))
-        return;
-
     cd = pa_sprintf_malloc("%s/snd/controlC%s", udev_get_dev_path(u->udev), path_get_card_id(d->path));
-    d->accessible = access(cd, W_OK) >= 0;
-    pa_log_info("%s is accessible: %s", cd, pa_yes_no(d->accessible));
+    accessible = access(cd, R_OK|W_OK) >= 0;
+    pa_log_debug("%s is accessible: %s", cd, pa_yes_no(accessible));
+
     pa_xfree(cd);
 
-    pa_card_suspend(card, !d->accessible, PA_SUSPEND_SESSION);
+    if (d->module == PA_INVALID_INDEX) {
+
+        /* If we are not loaded, try to load */
+
+        if (accessible) {
+            pa_module *m;
+            pa_bool_t busy;
+
+            /* Check if any of the PCM devices that belong to this
+             * card are currently busy. If they are, don't try to load
+             * right now, to make sure the probing phase can
+             * successfully complete. When the current user of the
+             * device closes it we will get another notification via
+             * inotify and can then recheck. */
+
+            busy = is_card_busy(path_get_card_id(d->path));
+            pa_log_debug("%s is busy: %s", d->path, pa_yes_no(busy));
+
+            if (!busy) {
+                pa_log_debug("Loading module-alsa-card with arguments '%s'", d->args);
+                m = pa_module_load(u->core, "module-alsa-card", d->args);
+
+                if (m) {
+                    d->module = m->index;
+                    pa_log_info("Card %s (%s) module loaded.", d->path, d->card_name);
+                } else
+                    pa_log_info("Card %s (%s) failed to load module.", d->path, d->card_name);
+            }
+        }
+
+    } else {
+
+        /* If we are already loaded update suspend status with
+         * accessible boolean */
+
+        if ((card = pa_namereg_get(u->core, d->card_name, PA_NAMEREG_CARD)))
+            pa_card_suspend(card, !accessible, PA_SUSPEND_SESSION);
+    }
 }
 
 static void card_changed(struct userdata *u, struct udev_device *dev) {
     struct device *d;
     const char *path;
     const char *t;
-    char *card_name, *args;
-    pa_module *m;
     char *n;
 
     pa_assert(u);
@@ -135,44 +274,33 @@ static void card_changed(struct userdata *u, struct udev_device *dev) {
         return;
     }
 
+    d = pa_xnew0(struct device, 1);
+    d->path = pa_xstrdup(path);
+    d->module = PA_INVALID_INDEX;
+
     if (!(t = udev_device_get_property_value(dev, "PULSE_NAME")))
         if (!(t = udev_device_get_property_value(dev, "ID_ID")))
             if (!(t = udev_device_get_property_value(dev, "ID_PATH")))
                 t = path_get_card_id(path);
 
     n = pa_namereg_make_valid_name(t);
+    d->card_name = pa_sprintf_malloc("alsa_card.%s", n);
+    d->args = pa_sprintf_malloc("device_id=\"%s\" "
+                                "name=\"%s\" "
+                                "card_name=\"%s\" "
+                                "tsched=%s "
+                                "ignore_dB=%s "
+                                "card_properties=\"module-udev-detect.discovered=1\"",
+                                path_get_card_id(path),
+                                n,
+                                d->card_name,
+                                pa_yes_no(u->use_tsched),
+                                pa_yes_no(u->ignore_dB));
+    pa_xfree(n);
 
-    card_name = pa_sprintf_malloc("alsa_card.%s", n);
-    args = pa_sprintf_malloc("device_id=\"%s\" "
-                             "name=\"%s\" "
-                             "card_name=\"%s\" "
-                             "tsched=%s "
-                             "ignore_dB=%s "
-                             "card_properties=\"module-udev-detect.discovered=1\"",
-                             path_get_card_id(path),
-                             n,
-                             card_name,
-                             pa_yes_no(u->use_tsched),
-                             pa_yes_no(u->ignore_dB));
-
-    pa_log_debug("Loading module-alsa-card with arguments '%s'", args);
-    m = pa_module_load(u->core, "module-alsa-card", args);
-    pa_xfree(args);
-
-    if (m) {
-        pa_log_info("Card %s (%s) added.", path, n);
-
-        d = pa_xnew(struct device, 1);
-        d->path = pa_xstrdup(path);
-        d->card_name = card_name;
-        d->module = m->index;
-        d->accessible = TRUE;
-
-        pa_hashmap_put(u->devices, d->path, d);
-    } else
-        pa_xfree(card_name);
+    pa_hashmap_put(u->devices, d->path, d);
 
-    pa_xfree(n);
+    verify_access(u, d);
 }
 
 static void remove_card(struct userdata *u, struct udev_device *dev) {
@@ -185,7 +313,10 @@ static void remove_card(struct userdata *u, struct udev_device *dev) {
         return;
 
     pa_log_info("Card %s removed.", d->path);
-    pa_module_unload_request_by_index(u->core, d->module, TRUE);
+
+    if (d->module != PA_INVALID_INDEX)
+        pa_module_unload_request_by_index(u->core, d->module, TRUE);
+
     device_free(d);
 }
 
@@ -262,6 +393,34 @@ fail:
     u->udev_io = NULL;
 }
 
+static pa_bool_t pcm_node_belongs_to_device(
+        struct device *d,
+        const char *node) {
+
+    char *cd;
+    pa_bool_t b;
+
+    cd = pa_sprintf_malloc("pcmC%sD", path_get_card_id(d->path));
+    b = pa_startswith(node, cd);
+    pa_xfree(cd);
+
+    return b;
+}
+
+static pa_bool_t control_node_belongs_to_device(
+        struct device *d,
+        const char *node) {
+
+    char *cd;
+    pa_bool_t b;
+
+    cd = pa_sprintf_malloc("controlC%s", path_get_card_id(d->path));
+    b = pa_streq(node, cd);
+    pa_xfree(cd);
+
+    return b;
+}
+
 static void inotify_cb(
         pa_mainloop_api*a,
         pa_io_event* e,
@@ -275,10 +434,13 @@ static void inotify_cb(
     } buf;
     struct userdata *u = userdata;
     static int type = 0;
-    pa_bool_t verify = FALSE, deleted = FALSE;
+    pa_bool_t deleted = FALSE;
+    struct device *d;
+    void *state;
 
     for (;;) {
         ssize_t r;
+        struct inotify_event *event;
 
         pa_zero(buf);
         if ((r = pa_read(fd, &buf, sizeof(buf), &type)) <= 0) {
@@ -290,22 +452,51 @@ static void inotify_cb(
             goto fail;
         }
 
-        if ((buf.e.mask & IN_CLOSE_WRITE) && pa_startswith(buf.e.name, "pcmC"))
-            verify = TRUE;
-
-        if ((buf.e.mask & (IN_DELETE_SELF|IN_MOVE_SELF)))
-            deleted = TRUE;
+        event = &buf.e;
+        while (r > 0) {
+            size_t len;
+
+            if ((size_t) r < sizeof(struct inotify_event)) {
+                pa_log("read() too short.");
+                goto fail;
+            }
+
+            len = sizeof(struct inotify_event) + event->len;
+
+            if ((size_t) r < len) {
+                pa_log("Payload missing.");
+                goto fail;
+            }
+
+            /* From udev we get the guarantee that the control
+             * device's ACL is changed last. To avoid races when ACLs
+             * are changed we hence watch only the control device */
+            if (((event->mask & IN_ATTRIB) && pa_startswith(event->name, "controlC")))
+                PA_HASHMAP_FOREACH(d, u->devices, state)
+                    if (control_node_belongs_to_device(d, event->name))
+                        d->need_verify = TRUE;
+
+            /* ALSA doesn't really give us any guarantee on the closing
+             * order, so let's simply hope */
+            if (((event->mask & IN_CLOSE_WRITE) && pa_startswith(event->name, "pcmC")))
+                PA_HASHMAP_FOREACH(d, u->devices, state)
+                    if (pcm_node_belongs_to_device(d, event->name))
+                        d->need_verify = TRUE;
+
+            /* /dev/snd/ might have been removed */
+            if ((event->mask & (IN_DELETE_SELF|IN_MOVE_SELF)))
+                deleted = TRUE;
+
+            event = (struct inotify_event*) ((uint8_t*) event + len);
+            r -= len;
+        }
     }
 
-    if (verify) {
-        struct device *d;
-        void *state;
-
-        pa_log_debug("Verifying access.");
-
-        PA_HASHMAP_FOREACH(d, u->devices, state)
+    PA_HASHMAP_FOREACH(d, u->devices, state)
+        if (d->need_verify) {
+            d->need_verify = FALSE;
             verify_access(u, d);
-    }
+        }
 
     if (!deleted)
         return;
@@ -335,7 +526,7 @@ static int setup_inotify(struct userdata *u) {
     }
 
     dev_snd = pa_sprintf_malloc("%s/snd", udev_get_dev_path(u->udev));
-    r = inotify_add_watch(u->inotify_fd, dev_snd, IN_CLOSE_WRITE|IN_DELETE_SELF|IN_MOVE_SELF);
+    r = inotify_add_watch(u->inotify_fd, dev_snd, IN_ATTRIB|IN_CLOSE_WRITE|IN_DELETE_SELF|IN_MOVE_SELF);
     pa_xfree(dev_snd);
 
     if (r < 0) {
@@ -449,7 +640,7 @@ int pa__init(pa_module *m) {
 
     udev_enumerate_unref(enumerate);
 
-    pa_log_info("Loaded %u modules.", pa_hashmap_size(u->devices));
+    pa_log_info("Found %u cards.", pa_hashmap_size(u->devices));
 
     pa_modargs_free(ma);
 
diff --git a/src/pulse/glib-mainloop.h b/src/pulse/glib-mainloop.h
index 189513a8..67aba27d 100644
--- a/src/pulse/glib-mainloop.h
+++ b/src/pulse/glib-mainloop.h
@@ -56,7 +56,9 @@ pa_glib_mainloop *pa_glib_mainloop_new(GMainContext *c);
 /** Free the GLIB main loop object */
 void pa_glib_mainloop_free(pa_glib_mainloop* g);
 
-/** Return the abstract main loop API vtable for the GLIB main loop object */
+/** Return the abstract main loop API vtable for the GLIB main loop
+    object. No need of freeing the API as it is owned by the loop and
+    it is destroyed when this dies */
 pa_mainloop_api* pa_glib_mainloop_get_api(pa_glib_mainloop *g);
 
 PA_C_DECL_END
diff --git a/src/pulse/mainloop.h b/src/pulse/mainloop.h
index 4a83ebe8..63abd588 100644
--- a/src/pulse/mainloop.h
+++ b/src/pulse/mainloop.h
@@ -108,7 +108,9 @@ int pa_mainloop_iterate(pa_mainloop *m, int block, int *retval);
 /** Run unlimited iterations of the main loop object until the main loop's quit() routine is called. */
 int pa_mainloop_run(pa_mainloop *m, int *retval);
 
-/** Return the abstract main loop abstraction layer vtable for this main loop. */
+/** Return the abstract main loop abstraction layer vtable for this
+    main loop. No need of freeing the API as it is owned by the loop
+    and it is destroyed when this dies */
 pa_mainloop_api* pa_mainloop_get_api(pa_mainloop*m);
 
 /** Shutdown the main loop */
diff --git a/src/pulse/sample.c b/src/pulse/sample.c
index d5d38eda..9698d8a5 100644
--- a/src/pulse/sample.c
+++ b/src/pulse/sample.c
@@ -36,28 +36,27 @@
 
 #include "sample.h"
 
-size_t pa_sample_size_of_format(pa_sample_format_t f) {
-
-    static const size_t table[] = {
-        [PA_SAMPLE_U8] = 1,
-        [PA_SAMPLE_ULAW] = 1,
-        [PA_SAMPLE_ALAW] = 1,
-        [PA_SAMPLE_S16LE] = 2,
-        [PA_SAMPLE_S16BE] = 2,
-        [PA_SAMPLE_FLOAT32LE] = 4,
-        [PA_SAMPLE_FLOAT32BE] = 4,
-        [PA_SAMPLE_S32LE] = 4,
-        [PA_SAMPLE_S32BE] = 4,
-        [PA_SAMPLE_S24LE] = 3,
-        [PA_SAMPLE_S24BE] = 3,
-        [PA_SAMPLE_S24_32LE] = 4,
-        [PA_SAMPLE_S24_32BE] = 4
-    };
+static const size_t size_table[] = {
+    [PA_SAMPLE_U8] = 1,
+    [PA_SAMPLE_ULAW] = 1,
+    [PA_SAMPLE_ALAW] = 1,
+    [PA_SAMPLE_S16LE] = 2,
+    [PA_SAMPLE_S16BE] = 2,
+    [PA_SAMPLE_FLOAT32LE] = 4,
+    [PA_SAMPLE_FLOAT32BE] = 4,
+    [PA_SAMPLE_S32LE] = 4,
+    [PA_SAMPLE_S32BE] = 4,
+    [PA_SAMPLE_S24LE] = 3,
+    [PA_SAMPLE_S24BE] = 3,
+    [PA_SAMPLE_S24_32LE] = 4,
+    [PA_SAMPLE_S24_32BE] = 4
+};
 
+size_t pa_sample_size_of_format(pa_sample_format_t f) {
     pa_assert(f >= 0);
     pa_assert(f < PA_SAMPLE_MAX);
 
-    return table[f];
+    return size_table[f];
 }
 
 size_t pa_sample_size(const pa_sample_spec *spec) {
@@ -65,35 +64,35 @@ size_t pa_sample_size(const pa_sample_spec *spec) {
     pa_assert(spec);
     pa_return_val_if_fail(pa_sample_spec_valid(spec), 0);
 
-    return pa_sample_size_of_format(spec->format);
+    return size_table[spec->format];
 }
 
 size_t pa_frame_size(const pa_sample_spec *spec) {
     pa_assert(spec);
     pa_return_val_if_fail(pa_sample_spec_valid(spec), 0);
 
-    return pa_sample_size(spec) * spec->channels;
+    return size_table[spec->format] * spec->channels;
 }
 
 size_t pa_bytes_per_second(const pa_sample_spec *spec) {
     pa_assert(spec);
     pa_return_val_if_fail(pa_sample_spec_valid(spec), 0);
 
-    return spec->rate*pa_frame_size(spec);
+    return spec->rate * size_table[spec->format] * spec->channels;
 }
 
 pa_usec_t pa_bytes_to_usec(uint64_t length, const pa_sample_spec *spec) {
     pa_assert(spec);
     pa_return_val_if_fail(pa_sample_spec_valid(spec), 0);
 
-    return (((pa_usec_t) (length / pa_frame_size(spec)) * PA_USEC_PER_SEC) / spec->rate);
+    return (((pa_usec_t) (length / (size_table[spec->format] * spec->channels)) * PA_USEC_PER_SEC) / spec->rate);
 }
 
 size_t pa_usec_to_bytes(pa_usec_t t, const pa_sample_spec *spec) {
     pa_assert(spec);
     pa_return_val_if_fail(pa_sample_spec_valid(spec), 0);
 
-    return (size_t) (((t * spec->rate) / PA_USEC_PER_SEC)) * pa_frame_size(spec);
+    return (size_t) (((t * spec->rate) / PA_USEC_PER_SEC)) * (size_table[spec->format] * spec->channels);
 }
 
 pa_sample_spec* pa_sample_spec_init(pa_sample_spec *spec) {
@@ -109,12 +108,12 @@ pa_sample_spec* pa_sample_spec_init(pa_sample_spec *spec) {
 int pa_sample_spec_valid(const pa_sample_spec *spec) {
     pa_assert(spec);
 
-    if (spec->rate <= 0 ||
+    if (PA_UNLIKELY (spec->rate <= 0 ||
         spec->rate > PA_RATE_MAX ||
         spec->channels <= 0 ||
         spec->channels > PA_CHANNELS_MAX ||
         spec->format >= PA_SAMPLE_MAX ||
-        spec->format < 0)
+        spec->format < 0))
         return 0;
 
     return 1;
diff --git a/src/pulse/thread-mainloop.h b/src/pulse/thread-mainloop.h
index e847070d..2cf496e1 100644
--- a/src/pulse/thread-mainloop.h
+++ b/src/pulse/thread-mainloop.h
@@ -299,7 +299,9 @@ void pa_threaded_mainloop_accept(pa_threaded_mainloop *m);
 /** Return the return value as specified with the main loop's quit() routine. */
 int pa_threaded_mainloop_get_retval(pa_threaded_mainloop *m);
 
-/** Return the abstract main loop abstraction layer vtable for this main loop. */
+/** Return the abstract main loop abstraction layer vtable for this
+    main loop. No need of freeing the API as it is owned by the loop
+    and it is destroyed when this dies */
 pa_mainloop_api* pa_threaded_mainloop_get_api(pa_threaded_mainloop*m);
 
 /** Returns non-zero when called from withing the event loop thread. \since 0.9.7 */
diff --git a/src/pulse/util.c b/src/pulse/util.c
index 6f1e40a9..9440f5de 100644
--- a/src/pulse/util.c
+++ b/src/pulse/util.c
@@ -61,38 +61,40 @@
 #include <pulsecore/log.h>
 #include <pulsecore/core-util.h>
 #include <pulsecore/macro.h>
+#include <pulsecore/usergroup.h>
 
 #include "util.h"
 
 char *pa_get_user_name(char *s, size_t l) {
     const char *p;
+    char *name = NULL;
+#ifdef OS_IS_WIN32
     char buf[1024];
+#endif
 
 #ifdef HAVE_PWD_H
-    struct passwd pw, *r;
+    struct passwd *r;
 #endif
 
     pa_assert(s);
     pa_assert(l > 0);
 
-    if (!(p = (getuid() == 0 ? "root" : NULL)) &&
-        !(p = getenv("USER")) &&
-        !(p = getenv("LOGNAME")) &&
-        !(p = getenv("USERNAME"))) {
+    if ((p = (getuid() == 0 ? "root" : NULL)) ||
+        (p = getenv("USER")) ||
+        (p = getenv("LOGNAME")) ||
+        (p = getenv("USERNAME")))
+    {
+        name = pa_strlcpy(s, p, l);
+    } else {
 #ifdef HAVE_PWD_H
 
-#ifdef HAVE_GETPWUID_R
-        if (getpwuid_r(getuid(), &pw, buf, sizeof(buf), &r) != 0 || !r) {
-#else
-        /* XXX Not thread-safe, but needed on OSes (e.g. FreeBSD 4.X)
-            * that do not support getpwuid_r. */
-        if ((r = getpwuid(getuid())) == NULL) {
-#endif
+        if ((r = pa_getpwuid_malloc(getuid())) == NULL) {
             pa_snprintf(s, l, "%lu", (unsigned long) getuid());
             return s;
         }
 
-        p = r->pw_name;
+        name = pa_strlcpy(s, r->pw_name, l);
+        pa_getpwuid_free(r);
 
 #elif defined(OS_IS_WIN32) /* HAVE_PWD_H */
         DWORD size = sizeof(buf);
@@ -102,7 +104,7 @@ char *pa_get_user_name(char *s, size_t l) {
             return NULL;
         }
 
-        p = buf;
+        name = pa_strlcpy(s, buf, l);
 
 #else /* HAVE_PWD_H */
 
@@ -110,7 +112,7 @@ char *pa_get_user_name(char *s, size_t l) {
 #endif /* HAVE_PWD_H */
     }
 
-    return pa_strlcpy(s, p, l);
+    return name;
 }
 
 char *pa_get_host_name(char *s, size_t l) {
@@ -126,11 +128,10 @@ char *pa_get_host_name(char *s, size_t l) {
 }
 
 char *pa_get_home_dir(char *s, size_t l) {
-    char *e;
+    char *e, *dir;
 
 #ifdef HAVE_PWD_H
-    char buf[1024];
-    struct passwd pw, *r;
+    struct passwd *r;
 #endif
 
     pa_assert(s);
@@ -143,22 +144,19 @@ char *pa_get_home_dir(char *s, size_t l) {
         return pa_strlcpy(s, e, l);
 
 #ifdef HAVE_PWD_H
-
     errno = 0;
-#ifdef HAVE_GETPWUID_R
-    if (getpwuid_r(getuid(), &pw, buf, sizeof(buf), &r) != 0 || !r) {
-#else
-    /* XXX Not thread-safe, but needed on OSes (e.g. FreeBSD 4.X)
-        * that do not support getpwuid_r. */
-    if ((r = getpwuid(getuid())) == NULL) {
-#endif
+    if ((r = pa_getpwuid_malloc(getuid())) == NULL) {
         if (!errno)
             errno = ENOENT;
 
         return NULL;
     }
 
-    return pa_strlcpy(s, r->pw_dir, l);
+    dir = pa_strlcpy(s, r->pw_dir, l);
+
+    pa_getpwuid_free(r);
+
+    return dir;
 #else /* HAVE_PWD_H */
 
     errno = ENOENT;
diff --git a/src/pulsecore/core-util.c b/src/pulsecore/core-util.c
index e83563f9..63751f53 100644
--- a/src/pulsecore/core-util.c
+++ b/src/pulsecore/core-util.c
@@ -115,6 +115,7 @@
 #include <pulsecore/macro.h>
 #include <pulsecore/thread.h>
 #include <pulsecore/strbuf.h>
+#include <pulsecore/usergroup.h>
 
 #include "core-util.h"
 
@@ -969,54 +970,24 @@ fail:
 
 /* Check whether the specified GID and the group name match */
 static int is_group(gid_t gid, const char *name) {
-    struct group group, *result = NULL;
-    long n;
-    void *data;
+    struct group *group = NULL;
     int r = -1;
 
-#ifdef HAVE_GETGRGID_R
-#ifdef _SC_GETGR_R_SIZE_MAX
-    n = sysconf(_SC_GETGR_R_SIZE_MAX);
-#else
-    n = -1;
-#endif
-    if (n <= 0)
-        n = 512;
-
-    data = pa_xmalloc((size_t) n);
-
     errno = 0;
-    if (getgrgid_r(gid, &group, data, (size_t) n, &result) < 0 || !result) {
-        pa_log("getgrgid_r(%u): %s", (unsigned) gid, pa_cstrerror(errno));
-
+    if (!(group = pa_getgrgid_malloc(gid)))
+    {
         if (!errno)
             errno = ENOENT;
 
-        goto finish;
-    }
-
-    r = strcmp(name, result->gr_name) == 0;
-
-finish:
-    pa_xfree(data);
-#else
-    /* XXX Not thread-safe, but needed on OSes (e.g. FreeBSD 4.X) that do not
-     * support getgrgid_r. */
-
-    errno = 0;
-    if (!(result = getgrgid(gid))) {
-        pa_log("getgrgid(%u): %s", gid, pa_cstrerror(errno));
-
-        if (!errno)
-            errno = ENOENT;
+        pa_log("pa_getgrgid_malloc(%u): %s", gid, pa_cstrerror(errno));
 
         goto finish;
     }
 
-    r = strcmp(name, result->gr_name) == 0;
+    r = strcmp(name, group->gr_name) == 0;
 
 finish:
-#endif
+    pa_getgrgid_free(group);
 
     return r;
 }
@@ -1065,38 +1036,12 @@ finish:
 
 /* Check whether the specifc user id is a member of the specified group */
 int pa_uid_in_group(uid_t uid, const char *name) {
-    char *g_buf, *p_buf;
-    long g_n, p_n;
-    struct group grbuf, *gr;
+    struct group *group = NULL;
     char **i;
     int r = -1;
 
-#ifdef _SC_GETGR_R_SIZE_MAX
-    g_n = sysconf(_SC_GETGR_R_SIZE_MAX);
-#else
-    g_n = -1;
-#endif
-    if (g_n <= 0)
-        g_n = 512;
-
-    g_buf = pa_xmalloc((size_t) g_n);
-
-#ifdef _SC_GETPW_R_SIZE_MAX
-    p_n = sysconf(_SC_GETPW_R_SIZE_MAX);
-#else
-    p_n = -1;
-#endif
-    if (p_n <= 0)
-        p_n = 512;
-
-    p_buf = pa_xmalloc((size_t) p_n);
-
     errno = 0;
-#ifdef HAVE_GETGRNAM_R
-    if (getgrnam_r(name, &grbuf, g_buf, (size_t) g_n, &gr) != 0 || !gr)
-#else
-    if (!(gr = getgrnam(name)))
-#endif
+    if (!(group = pa_getgrnam_malloc(name)))
     {
         if (!errno)
             errno = ENOENT;
@@ -1104,25 +1049,24 @@ int pa_uid_in_group(uid_t uid, const char *name) {
     }
 
     r = 0;
-    for (i = gr->gr_mem; *i; i++) {
-        struct passwd pwbuf, *pw;
+    for (i = group->gr_mem; *i; i++) {
+        struct passwd *pw = NULL;
 
-#ifdef HAVE_GETPWNAM_R
-        if (getpwnam_r(*i, &pwbuf, p_buf, (size_t) p_n, &pw) != 0 || !pw)
-#else
-        if (!(pw = getpwnam(*i)))
-#endif
+        errno = 0;
+        if (!(pw = pa_getpwnam_malloc(*i)))
             continue;
 
-        if (pw->pw_uid == uid) {
+        if (pw->pw_uid == uid)
             r = 1;
+
+        pa_getpwnam_free(pw);
+
+        if (r == 1)
             break;
-        }
     }
 
 finish:
-    pa_xfree(g_buf);
-    pa_xfree(p_buf);
+    pa_getgrnam_free(group);
 
     return r;
 }
@@ -1130,26 +1074,10 @@ finish:
 /* Get the GID of a gfiven group, return (gid_t) -1 on failure. */
 gid_t pa_get_gid_of_group(const char *name) {
     gid_t ret = (gid_t) -1;
-    char *g_buf;
-    long g_n;
-    struct group grbuf, *gr;
-
-#ifdef _SC_GETGR_R_SIZE_MAX
-    g_n = sysconf(_SC_GETGR_R_SIZE_MAX);
-#else
-    g_n = -1;
-#endif
-    if (g_n <= 0)
-        g_n = 512;
-
-    g_buf = pa_xmalloc((size_t) g_n);
+    struct group *gr = NULL;
 
     errno = 0;
-#ifdef HAVE_GETGRNAM_R
-    if (getgrnam_r(name, &grbuf, g_buf, (size_t) g_n, &gr) != 0 || !gr)
-#else
-    if (!(gr = getgrnam(name)))
-#endif
+    if (!(gr = pa_getgrnam_malloc(name)))
     {
         if (!errno)
             errno = ENOENT;
@@ -1159,7 +1087,7 @@ gid_t pa_get_gid_of_group(const char *name) {
     ret = gr->gr_gid;
 
 finish:
-    pa_xfree(g_buf);
+    pa_getgrnam_free(gr);
     return ret;
 }
 
@@ -2295,7 +2223,7 @@ int pa_close_all(int except_fd, ...) {
     va_end(ap);
 
     r = pa_close_allv(p);
-    free(p);
+    pa_xfree(p);
 
     return r;
 }
@@ -2890,3 +2818,22 @@ void pa_reset_personality(void) {
 #endif
 
 }
+
+#if defined(__linux__) && !defined(__OPTIMIZE__)
+
+pa_bool_t pa_run_from_build_tree(void) {
+    char *rp;
+    pa_bool_t b = FALSE;
+
+    /* We abuse __OPTIMIZE__ as a check whether we are a debug build
+     * or not. */
+
+    if ((rp = pa_readlink("/proc/self/exe"))) {
+        b = pa_startswith(rp, PA_BUILDDIR);
+        pa_xfree(rp);
+    }
+
+    return b;
+}
+
+#endif
diff --git a/src/pulsecore/core-util.h b/src/pulsecore/core-util.h
index 3db55106..8c13b535 100644
--- a/src/pulsecore/core-util.h
+++ b/src/pulsecore/core-util.h
@@ -250,4 +250,8 @@ size_t pa_pipe_buf(int fd);
 
 void pa_reset_personality(void);
 
+#if defined(__linux__) && !defined(__OPTIMIZE__)
+pa_bool_t pa_run_from_build_tree(void);
+#endif
+
 #endif
diff --git a/src/pulsecore/core.c b/src/pulsecore/core.c
index f5eb8352..f0726453 100644
--- a/src/pulsecore/core.c
+++ b/src/pulsecore/core.c
@@ -47,7 +47,7 @@
 
 #include "core.h"
 
-static PA_DEFINE_CHECK_TYPE(pa_core, pa_msgobject);
+PA_DEFINE_PUBLIC_CLASS(pa_core, pa_msgobject);
 
 static int core_process_msg(pa_msgobject *o, int code, void *userdata, int64_t offset, pa_memchunk *chunk) {
     pa_core *c = PA_CORE(o);
diff --git a/src/pulsecore/core.h b/src/pulsecore/core.h
index fc4eabfa..bfcea4f6 100644
--- a/src/pulsecore/core.h
+++ b/src/pulsecore/core.h
@@ -174,7 +174,7 @@ struct pa_core {
     pa_hook hooks[PA_CORE_HOOK_MAX];
 };
 
-PA_DECLARE_CLASS(pa_core);
+PA_DECLARE_PUBLIC_CLASS(pa_core);
 #define PA_CORE(o) pa_core_cast(o)
 
 enum {
diff --git a/src/pulsecore/cpu-arm.c b/src/pulsecore/cpu-arm.c
new file mode 100644
index 00000000..453b7848
--- /dev/null
+++ b/src/pulsecore/cpu-arm.c
@@ -0,0 +1,139 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk>
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2.1 of the License,
+  or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <stdint.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+
+#include <pulse/xmalloc.h>
+#include <pulsecore/log.h>
+
+#include "cpu-arm.h"
+
+#if defined (__arm__) && defined (__linux__)
+
+#define MAX_BUFFER  4096
+static char *
+get_cpuinfo_line (char *cpuinfo, const char *tag) {
+    char *line, *end, *colon;
+
+    if (!(line = strstr (cpuinfo, tag)))
+        return NULL;
+
+    if (!(end = strchr (line, '\n')))
+        return NULL;
+
+    if (!(colon = strchr (line, ':')))
+        return NULL;
+
+    if (++colon >= end)
+        return NULL;
+
+    return pa_xstrndup (colon, end - colon);
+}
+
+static char *get_cpuinfo(void) {
+    char *cpuinfo;
+    int n, fd;
+
+    cpuinfo = pa_xmalloc(MAX_BUFFER);
+
+    if ((fd = open("/proc/cpuinfo", O_RDONLY)) < 0) {
+        pa_xfree(cpuinfo);
+        return NULL;
+    }
+
+    if ((n = pa_read(fd, cpuinfo, MAX_BUFFER-1)) < 0) {
+        pa_xfree(cpuinfo);
+        pa_close(fd);
+        return NULL;
+    }
+    cpuinfo[n] = 0;
+    pa_close(fd);
+
+    return cpuinfo;
+}
+#endif /* defined (__arm__) && defined (__linux__) */
+
+void pa_cpu_init_arm (void) {
+#if defined (__arm__)
+#if defined (__linux__)
+    char *cpuinfo, *line;
+    int arch;
+    pa_cpu_arm_flag_t flags = 0;
+
+    /* We need to read the CPU flags from /proc/cpuinfo because there is no user
+     * space support to get the CPU features. This only works on linux AFAIK. */
+    if (!(cpuinfo = get_cpuinfo ())) {
+        pa_log ("Can't read cpuinfo");
+        return;
+    }
+
+    /* get the CPU architecture */
+    if ((line = get_cpuinfo_line (cpuinfo, "CPU architecture"))) {
+        arch = strtoul (line, NULL, 0);
+        if (arch >= 6)
+            flags |= PA_CPU_ARM_V6;
+        if (arch >= 7)
+            flags |= PA_CPU_ARM_V7;
+
+        pa_xfree(line);
+    }
+    /* get the CPU features */
+    if ((line = get_cpuinfo_line (cpuinfo, "Features"))) {
+        char *state = NULL, *current;
+
+        while ((current = pa_split_spaces (line, &state))) {
+            if (!strcmp (current, "vfp"))
+                flags |= PA_CPU_ARM_VFP;
+            else if (!strcmp (current, "edsp"))
+                flags |= PA_CPU_ARM_EDSP;
+            else if (!strcmp (current, "neon"))
+                flags |= PA_CPU_ARM_NEON;
+            else if (!strcmp (current, "vfpv3"))
+                flags |= PA_CPU_ARM_VFPV3;
+
+            pa_xfree(current);
+        }
+    }
+    pa_xfree(cpuinfo);
+
+    pa_log_info ("CPU flags: %s%s%s%s%s%s",
+          (flags & PA_CPU_ARM_V6) ? "V6 " : "",
+          (flags & PA_CPU_ARM_V7) ? "V7 " : "",
+          (flags & PA_CPU_ARM_VFP) ? "VFP " : "",
+          (flags & PA_CPU_ARM_EDSP) ? "EDSP " : "",
+          (flags & PA_CPU_ARM_NEON) ? "NEON " : "",
+          (flags & PA_CPU_ARM_VFPV3) ? "VFPV3 " : "");
+#else /* defined (__linux__) */
+    pa_log ("ARM cpu features not yet supported on this OS");
+#endif /* defined (__linux__) */
+
+    if (flags & PA_CPU_ARM_V6)
+        pa_volume_func_init_arm (flags);
+#endif /* defined (__arm__) */
+}
diff --git a/src/pulsecore/cpu-arm.h b/src/pulsecore/cpu-arm.h
new file mode 100644
index 00000000..a87cb63b
--- /dev/null
+++ b/src/pulsecore/cpu-arm.h
@@ -0,0 +1,42 @@
+#ifndef foocpuarmhfoo
+#define foocpuarmhfoo
+
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk>
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2.1 of the License,
+  or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#include <stdint.h>
+
+typedef enum pa_cpu_arm_flag {
+    PA_CPU_ARM_V6       = (1 << 0),
+    PA_CPU_ARM_V7       = (1 << 1),
+    PA_CPU_ARM_VFP      = (1 << 2),
+    PA_CPU_ARM_EDSP     = (1 << 3),
+    PA_CPU_ARM_NEON     = (1 << 4),
+    PA_CPU_ARM_VFPV3    = (1 << 5)
+} pa_cpu_arm_flag_t;
+
+void pa_cpu_init_arm (void);
+
+/* some optimized functions */
+void pa_volume_func_init_arm(pa_cpu_arm_flag_t flags);
+
+#endif /* foocpuarmhfoo */
diff --git a/src/pulsecore/cpu-x86.c b/src/pulsecore/cpu-x86.c
new file mode 100644
index 00000000..bc093ec0
--- /dev/null
+++ b/src/pulsecore/cpu-x86.c
@@ -0,0 +1,122 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk>
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2.1 of the License,
+  or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <stdint.h>
+
+#include <pulsecore/log.h>
+
+#include "cpu-x86.h"
+
+#if defined (__i386__) || defined (__amd64__)
+static void
+get_cpuid (uint32_t op, uint32_t *a, uint32_t *b, uint32_t *c, uint32_t *d)
+{
+    __asm__ __volatile__ (
+        "  push %%"PA_REG_b"   \n\t"
+        "  cpuid               \n\t"
+        "  mov %%ebx, %%esi    \n\t"
+        "  pop %%"PA_REG_b"    \n\t"
+
+        : "=a" (*a), "=S" (*b), "=c" (*c), "=d" (*d)
+        : "0" (op)
+    );
+}
+#endif
+
+void pa_cpu_init_x86 (void) {
+#if defined (__i386__) || defined (__amd64__)
+    uint32_t eax, ebx, ecx, edx;
+    uint32_t level;
+    pa_cpu_x86_flag_t flags = 0;
+
+    /* get standard level */
+    get_cpuid (0x00000000, &level, &ebx, &ecx, &edx);
+    if (level >= 1) {
+        get_cpuid (0x00000001, &eax, &ebx, &ecx, &edx);
+
+        if (edx & (1<<23))
+          flags |= PA_CPU_X86_MMX;
+
+        if (edx & (1<<25))
+          flags |= PA_CPU_X86_SSE;
+
+        if (edx & (1<<26))
+          flags |= PA_CPU_X86_SSE2;
+
+        if (ecx & (1<<0))
+          flags |= PA_CPU_X86_SSE3;
+
+        if (ecx & (1<<9))
+          flags |= PA_CPU_X86_SSSE3;
+
+        if (ecx & (1<<19))
+          flags |= PA_CPU_X86_SSE4_1;
+
+        if (ecx & (1<<20))
+          flags |= PA_CPU_X86_SSE4_2;
+    }
+
+    /* get extended level */
+    get_cpuid (0x80000000, &level, &ebx, &ecx, &edx);
+    if (level >= 0x80000001) {
+        get_cpuid (0x80000001, &eax, &ebx, &ecx, &edx);
+
+        if (edx & (1<<22))
+          flags |= PA_CPU_X86_MMXEXT;
+
+        if (edx & (1<<23))
+          flags |= PA_CPU_X86_MMX;
+
+        if (edx & (1<<30))
+          flags |= PA_CPU_X86_3DNOWEXT;
+
+        if (edx & (1<<31))
+          flags |= PA_CPU_X86_3DNOW;
+    }
+
+    pa_log_info ("CPU flags: %s%s%s%s%s%s%s%s%s%s",
+    (flags & PA_CPU_X86_MMX) ? "MMX " : "",
+    (flags & PA_CPU_X86_SSE) ? "SSE " : "",
+    (flags & PA_CPU_X86_SSE2) ? "SSE2 " : "",
+    (flags & PA_CPU_X86_SSE3) ? "SSE3 " : "",
+    (flags & PA_CPU_X86_SSSE3) ? "SSSE3 " : "",
+    (flags & PA_CPU_X86_SSE4_1) ? "SSE4_1 " : "",
+    (flags & PA_CPU_X86_SSE4_2) ? "SSE4_2 " : "",
+    (flags & PA_CPU_X86_MMXEXT) ? "MMXEXT " : "",
+    (flags & PA_CPU_X86_3DNOW) ? "3DNOW " : "",
+    (flags & PA_CPU_X86_3DNOWEXT) ? "3DNOWEXT " : "");
+
+    /* activate various optimisations */
+    if (flags & PA_CPU_X86_MMX) {
+        pa_volume_func_init_mmx (flags);
+        pa_remap_func_init_mmx (flags);
+    }
+
+    if (flags & PA_CPU_X86_SSE)
+        pa_volume_func_init_sse (flags);
+
+#endif /* defined (__i386__) || defined (__amd64__) */
+}
diff --git a/src/pulsecore/cpu-x86.h b/src/pulsecore/cpu-x86.h
new file mode 100644
index 00000000..f3f9e56e
--- /dev/null
+++ b/src/pulsecore/cpu-x86.h
@@ -0,0 +1,68 @@
+#ifndef foocpux86hfoo
+#define foocpux86hfoo
+
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk>
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2.1 of the License,
+  or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#include <stdint.h>
+
+typedef enum pa_cpu_x86_flag {
+    PA_CPU_X86_MMX       = (1 << 0),
+    PA_CPU_X86_MMXEXT    = (1 << 1),
+    PA_CPU_X86_SSE       = (1 << 2),
+    PA_CPU_X86_SSE2      = (1 << 3),
+    PA_CPU_X86_SSE3      = (1 << 4),
+    PA_CPU_X86_SSSE3     = (1 << 5),
+    PA_CPU_X86_SSE4_1    = (1 << 6),
+    PA_CPU_X86_SSE4_2    = (1 << 7),
+    PA_CPU_X86_3DNOW     = (1 << 8),
+    PA_CPU_X86_3DNOWEXT  = (1 << 9)
+} pa_cpu_x86_flag_t;
+
+void pa_cpu_init_x86 (void);
+
+
+#if defined (__i386__)
+typedef int32_t pa_reg_x86;
+#define PA_REG_a "eax"
+#define PA_REG_b "ebx"
+#define PA_REG_c "ecx"
+#define PA_REG_d "edx"
+#define PA_REG_D "edi"
+#define PA_REG_S "esi"
+#elif defined (__amd64__)
+typedef int64_t pa_reg_x86;
+#define PA_REG_a "rax"
+#define PA_REG_b "rbx"
+#define PA_REG_c "rcx"
+#define PA_REG_d "rdx"
+#define PA_REG_D "rdi"
+#define PA_REG_S "rsi"
+#endif
+
+/* some optimized functions */
+void pa_volume_func_init_mmx(pa_cpu_x86_flag_t flags);
+void pa_volume_func_init_sse(pa_cpu_x86_flag_t flags);
+
+void pa_remap_func_init_mmx(pa_cpu_x86_flag_t flags);
+
+#endif /* foocpux86hfoo */
diff --git a/src/pulsecore/msgobject.c b/src/pulsecore/msgobject.c
index 6a2a612d..075a28c5 100644
--- a/src/pulsecore/msgobject.c
+++ b/src/pulsecore/msgobject.c
@@ -26,22 +26,22 @@
 
 #include "msgobject.h"
 
-PA_DEFINE_CHECK_TYPE(pa_msgobject, pa_object);
+PA_DEFINE_PUBLIC_CLASS(pa_msgobject, pa_object);
 
-pa_msgobject *pa_msgobject_new_internal(size_t size, const char *type_name, int (*check_type)(const char *type_name)) {
+pa_msgobject *pa_msgobject_new_internal(size_t size, const char *type_id, pa_bool_t (*check_type)(const char *type_name)) {
     pa_msgobject *o;
 
     pa_assert(size > sizeof(pa_msgobject));
-    pa_assert(type_name);
+    pa_assert(type_id);
 
     if (!check_type)
         check_type = pa_msgobject_check_type;
 
-    pa_assert(check_type(type_name));
-    pa_assert(check_type("pa_object"));
-    pa_assert(check_type("pa_msgobject"));
+    pa_assert(check_type(type_id));
+    pa_assert(check_type(pa_object_type_id));
+    pa_assert(check_type(pa_msgobject_type_id));
 
-    o = PA_MSGOBJECT(pa_object_new_internal(size, type_name, check_type));
+    o = PA_MSGOBJECT(pa_object_new_internal(size, type_id, check_type));
     o->process_msg = NULL;
     return o;
 }
diff --git a/src/pulsecore/msgobject.h b/src/pulsecore/msgobject.h
index a35a23b5..ee0ec1ed 100644
--- a/src/pulsecore/msgobject.h
+++ b/src/pulsecore/msgobject.h
@@ -38,15 +38,13 @@ struct pa_msgobject {
     int (*process_msg)(pa_msgobject *o, int code, void *userdata, int64_t offset, pa_memchunk *chunk);
 };
 
-pa_msgobject *pa_msgobject_new_internal(size_t size, const char *type_name, int (*check_type)(const char *type_name));
+pa_msgobject *pa_msgobject_new_internal(size_t size, const char *type_id, pa_bool_t (*check_type)(const char *type_name));
 
-int pa_msgobject_check_type(const char *type);
-
-#define pa_msgobject_new(type) ((type*) pa_msgobject_new_internal(sizeof(type), #type, type##_check_type))
+#define pa_msgobject_new(type) ((type*) pa_msgobject_new_internal(sizeof(type), type##_type_id, type##_check_type))
 #define pa_msgobject_free ((void (*) (pa_msgobject* o)) pa_object_free)
 
 #define PA_MSGOBJECT(o) pa_msgobject_cast(o)
 
-PA_DECLARE_CLASS(pa_msgobject);
+PA_DECLARE_PUBLIC_CLASS(pa_msgobject);
 
 #endif
diff --git a/src/pulsecore/object.c b/src/pulsecore/object.c
index f3ead9c5..099d50d9 100644
--- a/src/pulsecore/object.c
+++ b/src/pulsecore/object.c
@@ -28,21 +28,23 @@
 
 #include "object.h"
 
-pa_object *pa_object_new_internal(size_t size, const char *type_name, int (*check_type)(const char *type_name)) {
+const char pa_object_type_id[] = "pa_object";
+
+pa_object *pa_object_new_internal(size_t size, const char *type_id, pa_bool_t (*check_type)(const char *type_id)) {
     pa_object *o;
 
     pa_assert(size > sizeof(pa_object));
-    pa_assert(type_name);
+    pa_assert(type_id);
 
     if (!check_type)
         check_type = pa_object_check_type;
 
-    pa_assert(check_type(type_name));
-    pa_assert(check_type("pa_object"));
+    pa_assert(check_type(type_id));
+    pa_assert(check_type(pa_object_type_id));
 
     o = pa_xmalloc(size);
     PA_REFCNT_INIT(o);
-    o->type_name = type_name;
+    o->type_id = type_id;
     o->free = pa_object_free;
     o->check_type = check_type;
 
@@ -65,8 +67,8 @@ void pa_object_unref(pa_object *o) {
     }
 }
 
-int pa_object_check_type(const char *type_name) {
-    pa_assert(type_name);
+pa_bool_t pa_object_check_type(const char *type_id) {
+    pa_assert(type_id);
 
-    return pa_streq(type_name, "pa_object");
+    return type_id == pa_object_type_id;
 }
diff --git a/src/pulsecore/object.h b/src/pulsecore/object.h
index 43e79327..4c120cd5 100644
--- a/src/pulsecore/object.h
+++ b/src/pulsecore/object.h
@@ -34,21 +34,23 @@ typedef struct pa_object pa_object;
 
 struct pa_object {
     PA_REFCNT_DECLARE;
-    const char *type_name;
+    const char *type_id;
     void (*free)(pa_object *o);
-    int (*check_type)(const char *type_name);
+    pa_bool_t (*check_type)(const char *type_name);
 };
 
-pa_object *pa_object_new_internal(size_t size, const char *type_name, int (*check_type)(const char *type_name));
-#define pa_object_new(type) ((type*) pa_object_new_internal(sizeof(type), #type, type##_check_type)
+pa_object *pa_object_new_internal(size_t size, const char *type_id, pa_bool_t (*check_type)(const char *type_id));
+#define pa_object_new(type) ((type*) pa_object_new_internal(sizeof(type), type##_type_id, type##_check_type)
 
 #define pa_object_free ((void (*) (pa_object* _obj)) pa_xfree)
 
-int pa_object_check_type(const char *type);
+pa_bool_t pa_object_check_type(const char *type_id);
 
-static inline int pa_object_isinstance(void *o) {
+extern const char pa_object_type_id[];
+
+static inline pa_bool_t pa_object_isinstance(void *o) {
     pa_object *obj = (pa_object*) o;
-    return obj ? obj->check_type("pa_object") : 0;
+    return obj ? obj->check_type(pa_object_type_id) : TRUE;
 }
 
 pa_object *pa_object_ref(pa_object *o);
@@ -60,7 +62,7 @@ static inline int pa_object_refcnt(pa_object *o) {
 
 static inline pa_object* pa_object_cast(void *o) {
     pa_object *obj = (pa_object*) o;
-    pa_assert(!obj || obj->check_type("pa_object"));
+    pa_assert(!obj || obj->check_type(pa_object_type_id));
     return obj;
 }
 
@@ -68,10 +70,10 @@ static inline pa_object* pa_object_cast(void *o) {
 
 #define PA_OBJECT(o) pa_object_cast(o)
 
-#define PA_DECLARE_CLASS(c)                                             \
-    static inline int c##_isinstance(void *o) {                         \
+#define PA_DECLARE_CLASS_COMMON(c)                                      \
+    static inline pa_bool_t c##_isinstance(void *o) {                   \
         pa_object *obj = (pa_object*) o;                                \
-        return obj ? obj->check_type(#c) : 1;                           \
+        return obj ? obj->check_type(c##_type_id) : TRUE;               \
     }                                                                   \
     static inline c* c##_cast(void *o) {                                \
         pa_assert(c##_isinstance(o));                                   \
@@ -91,12 +93,27 @@ static inline pa_object* pa_object_cast(void *o) {
     }                                                                   \
     struct __stupid_useless_struct_to_allow_trailing_semicolon
 
-#define PA_DEFINE_CHECK_TYPE(c, parent)                                 \
-    int c##_check_type(const char *type) {                              \
-        pa_assert(type);                                                \
-        if (strcmp(type, #c) == 0)                                      \
-            return 1;                                                   \
-        return parent##_check_type(type);                               \
+#define PA_DECLARE_PUBLIC_CLASS(c)                                      \
+    extern const char c##_type_id[];                                    \
+    PA_DECLARE_CLASS_COMMON(c);                                         \
+    pa_bool_t c##_check_type(const char *type_id)
+
+#define PA_DEFINE_PUBLIC_CLASS(c, parent)                               \
+    const char c##_type_id[] = #c;                                      \
+    pa_bool_t c##_check_type(const char *type_id) {                     \
+        if (type_id == c##_type_id)                                     \
+            return TRUE;                                                \
+        return parent##_check_type(type_id);                            \
+    }                                                                   \
+    struct __stupid_useless_struct_to_allow_trailing_semicolon
+
+#define PA_DEFINE_PRIVATE_CLASS(c, parent)                              \
+    static const char c##_type_id[] = #c;                               \
+    PA_DECLARE_CLASS_COMMON(c);                                         \
+    static pa_bool_t c##_check_type(const char *type_id) {              \
+        if (type_id == c##_type_id)                                     \
+            return TRUE;                                                \
+        return parent##_check_type(type_id);                            \
     }                                                                   \
     struct __stupid_useless_struct_to_allow_trailing_semicolon
 
diff --git a/src/pulsecore/play-memblockq.c b/src/pulsecore/play-memblockq.c
index fceb2ca1..b0d76993 100644
--- a/src/pulsecore/play-memblockq.c
+++ b/src/pulsecore/play-memblockq.c
@@ -47,9 +47,8 @@ enum {
     MEMBLOCKQ_STREAM_MESSAGE_UNLINK,
 };
 
-PA_DECLARE_CLASS(memblockq_stream);
+PA_DEFINE_PRIVATE_CLASS(memblockq_stream, pa_msgobject);
 #define MEMBLOCKQ_STREAM(o) (memblockq_stream_cast(o))
-static PA_DEFINE_CHECK_TYPE(memblockq_stream, pa_msgobject);
 
 static void memblockq_stream_unlink(memblockq_stream *u) {
     pa_assert(u);
diff --git a/src/pulsecore/protocol-esound.c b/src/pulsecore/protocol-esound.c
index f64552aa..cfbaee6f 100644
--- a/src/pulsecore/protocol-esound.c
+++ b/src/pulsecore/protocol-esound.c
@@ -120,9 +120,8 @@ typedef struct connection {
     pa_time_event *auth_timeout_event;
 } connection;
 
-PA_DECLARE_CLASS(connection);
+PA_DEFINE_PRIVATE_CLASS(connection, pa_msgobject);
 #define CONNECTION(o) (connection_cast(o))
-static PA_DEFINE_CHECK_TYPE(connection, pa_msgobject);
 
 struct pa_esound_protocol {
     PA_REFCNT_DECLARE;
diff --git a/src/pulsecore/protocol-native.c b/src/pulsecore/protocol-native.c
index b1285e15..6678d847 100644
--- a/src/pulsecore/protocol-native.c
+++ b/src/pulsecore/protocol-native.c
@@ -98,17 +98,15 @@ typedef struct record_stream {
     pa_usec_t current_source_latency;
 } record_stream;
 
-PA_DECLARE_CLASS(record_stream);
 #define RECORD_STREAM(o) (record_stream_cast(o))
-static PA_DEFINE_CHECK_TYPE(record_stream, pa_msgobject);
+PA_DEFINE_PRIVATE_CLASS(record_stream, pa_msgobject);
 
 typedef struct output_stream {
     pa_msgobject parent;
 } output_stream;
 
-PA_DECLARE_CLASS(output_stream);
 #define OUTPUT_STREAM(o) (output_stream_cast(o))
-static PA_DEFINE_CHECK_TYPE(output_stream, pa_msgobject);
+PA_DEFINE_PRIVATE_CLASS(output_stream, pa_msgobject);
 
 typedef struct playback_stream {
     output_stream parent;
@@ -138,9 +136,8 @@ typedef struct playback_stream {
     uint64_t playing_for, underrun_for;
 } playback_stream;
 
-PA_DECLARE_CLASS(playback_stream);
 #define PLAYBACK_STREAM(o) (playback_stream_cast(o))
-static PA_DEFINE_CHECK_TYPE(playback_stream, output_stream);
+PA_DEFINE_PRIVATE_CLASS(playback_stream, output_stream);
 
 typedef struct upload_stream {
     output_stream parent;
@@ -156,9 +153,8 @@ typedef struct upload_stream {
     pa_proplist *proplist;
 } upload_stream;
 
-PA_DECLARE_CLASS(upload_stream);
 #define UPLOAD_STREAM(o) (upload_stream_cast(o))
-static PA_DEFINE_CHECK_TYPE(upload_stream, output_stream);
+PA_DEFINE_PRIVATE_CLASS(upload_stream, output_stream);
 
 struct pa_native_connection {
     pa_msgobject parent;
@@ -176,9 +172,8 @@ struct pa_native_connection {
     pa_time_event *auth_timeout_event;
 };
 
-PA_DECLARE_CLASS(pa_native_connection);
 #define PA_NATIVE_CONNECTION(o) (pa_native_connection_cast(o))
-static PA_DEFINE_CHECK_TYPE(pa_native_connection, pa_msgobject);
+PA_DEFINE_PRIVATE_CLASS(pa_native_connection, pa_msgobject);
 
 struct pa_native_protocol {
     PA_REFCNT_DECLARE;
diff --git a/src/pulsecore/protocol-simple.c b/src/pulsecore/protocol-simple.c
index 776d74b6..95ec6ac8 100644
--- a/src/pulsecore/protocol-simple.c
+++ b/src/pulsecore/protocol-simple.c
@@ -69,9 +69,8 @@ typedef struct connection {
     } playback;
 } connection;
 
-PA_DECLARE_CLASS(connection);
+PA_DEFINE_PRIVATE_CLASS(connection, pa_msgobject);
 #define CONNECTION(o) (connection_cast(o))
-static PA_DEFINE_CHECK_TYPE(connection, pa_msgobject);
 
 struct pa_simple_protocol {
     PA_REFCNT_DECLARE;
diff --git a/src/pulsecore/remap.c b/src/pulsecore/remap.c
new file mode 100644
index 00000000..a0fc85b9
--- /dev/null
+++ b/src/pulsecore/remap.c
@@ -0,0 +1,204 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk.com>
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2.1 of the License,
+  or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <string.h>
+
+#include <pulse/sample.h>
+#include <pulsecore/log.h>
+#include <pulsecore/macro.h>
+
+#include "remap.h"
+
+static void remap_mono_to_stereo_c (pa_remap_t *m, void *dst, const void *src, unsigned n) {
+    unsigned i;
+
+    switch (*m->format) {
+        case PA_SAMPLE_FLOAT32NE:
+        {
+            float *d, *s;
+
+            d = (float *) dst;
+            s = (float *) src;
+
+            for (i = n >> 2; i; i--) {
+                d[0] = d[1] = s[0];
+                d[2] = d[3] = s[1];
+                d[4] = d[5] = s[2];
+                d[6] = d[7] = s[3];
+                s += 4;
+                d += 8;
+            }
+            for (i = n & 3; i; i--) {
+                d[0] = d[1] = s[0];
+                s++;
+                d += 2;
+            }
+            break;
+        }
+        case PA_SAMPLE_S16NE:
+        {
+            int16_t *d, *s;
+
+            d = (int16_t *) dst;
+            s = (int16_t *) src;
+
+            for (i = n >> 2; i; i--) {
+                d[0] = d[1] = s[0];
+                d[2] = d[3] = s[1];
+                d[4] = d[5] = s[2];
+                d[6] = d[7] = s[3];
+                s += 4;
+                d += 8;
+            }
+            for (i = n & 3; i; i--) {
+                d[0] = d[1] = s[0];
+                s++;
+                d += 2;
+            }
+            break;
+        }
+        default:
+            pa_assert_not_reached();
+    }
+}
+
+static void remap_channels_matrix_c (pa_remap_t *m, void *dst, const void *src, unsigned n) {
+    unsigned oc, ic, i;
+    unsigned n_ic, n_oc;
+
+    n_ic = m->i_ss->channels;
+    n_oc = m->o_ss->channels;
+
+    switch (*m->format) {
+        case PA_SAMPLE_FLOAT32NE:
+        {
+            float *d, *s;
+
+            memset(dst, 0, n * sizeof (float) * n_oc);
+
+            for (oc = 0; oc < n_oc; oc++) {
+
+                for (ic = 0; ic < n_ic; ic++) {
+                    float vol;
+
+                    vol = m->map_table_f[oc][ic];
+
+                    if (vol <= 0.0)
+                        continue;
+
+                    d = (float *)dst + oc;
+                    s = (float *)src + ic;
+
+                    if (vol >= 1.0) {
+                        for (i = n; i > 0; i--, s += n_ic, d += n_oc)
+                            *d += *s;
+                    } else {
+                        for (i = n; i > 0; i--, s += n_ic, d += n_oc)
+                            *d += *s * vol;
+                    }
+                }
+            }
+
+            break;
+        }
+        case PA_SAMPLE_S16NE:
+        {
+            int16_t *d, *s;
+
+            memset(dst, 0, n * sizeof (int16_t) * n_oc);
+
+            for (oc = 0; oc < n_oc; oc++) {
+
+                for (ic = 0; ic < n_ic; ic++) {
+                    int32_t vol;
+
+                    vol = m->map_table_i[oc][ic];
+
+                    if (vol <= 0)
+                        continue;
+
+                    d = (int16_t *)dst + oc;
+                    s = (int16_t *)src + ic;
+
+                    if (vol >= 0x10000) {
+                        for (i = n; i > 0; i--, s += n_ic, d += n_oc)
+                            *d += *s;
+                    } else {
+                        for (i = n; i > 0; i--, s += n_ic, d += n_oc)
+                            *d += (int16_t) (((int32_t)*s * vol) >> 16);
+                    }
+                }
+            }
+            break;
+        }
+        default:
+            pa_assert_not_reached();
+    }
+}
+
+/* set the function that will execute the remapping based on the matrices */
+static void init_remap_c (pa_remap_t *m) {
+    unsigned n_oc, n_ic;
+
+    n_oc = m->o_ss->channels;
+    n_ic = m->i_ss->channels;
+
+    /* find some common channel remappings, fall back to full matrix operation. */
+    if (n_ic == 1 && n_oc == 2 &&
+            m->map_table_f[0][0] >= 1.0 && m->map_table_f[1][0] >= 1.0) {
+        m->do_remap = (pa_do_remap_func_t) remap_mono_to_stereo_c;
+        pa_log_info("Using mono to stereo remapping");
+    } else {
+        m->do_remap = (pa_do_remap_func_t) remap_channels_matrix_c;
+        pa_log_info("Using generic matrix remapping");
+    }
+}
+
+
+/* default C implementation */
+static pa_init_remap_func_t remap_func = init_remap_c;
+
+void pa_init_remap (pa_remap_t *m) {
+    pa_assert (remap_func);
+
+    m->do_remap = NULL;
+
+    /* call the installed remap init function */
+    remap_func (m);
+
+    if (m->do_remap == NULL) {
+        /* nothing was installed, fallback to C version */
+        init_remap_c (m);
+    }
+}
+
+pa_init_remap_func_t pa_get_init_remap_func(void) {
+    return remap_func;
+}
+
+void pa_set_init_remap_func(pa_init_remap_func_t func) {
+    remap_func = func;
+}
diff --git a/src/pulsecore/remap.h b/src/pulsecore/remap.h
new file mode 100644
index 00000000..32a67cdd
--- /dev/null
+++ b/src/pulsecore/remap.h
@@ -0,0 +1,48 @@
+#ifndef fooremapfoo
+#define fooremapfoo
+
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk.com>
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2.1 of the License,
+  or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#include <pulse/sample.h>
+
+typedef struct pa_remap pa_remap_t;
+
+typedef void (*pa_do_remap_func_t) (pa_remap_t *m, void *d, const void *s, unsigned n);
+
+struct pa_remap {
+    pa_sample_format_t *format;
+    pa_sample_spec *i_ss, *o_ss;
+    float map_table_f[PA_CHANNELS_MAX][PA_CHANNELS_MAX];
+    int32_t map_table_i[PA_CHANNELS_MAX][PA_CHANNELS_MAX];
+    pa_do_remap_func_t do_remap;
+};
+
+void pa_init_remap (pa_remap_t *m);
+
+/* custom installation of init functions */
+typedef void (*pa_init_remap_func_t) (pa_remap_t *m);
+
+pa_init_remap_func_t pa_get_init_remap_func(void);
+void pa_set_init_remap_func(pa_init_remap_func_t func);
+
+#endif /* fooremapfoo */
diff --git a/src/pulsecore/remap_mmx.c b/src/pulsecore/remap_mmx.c
new file mode 100644
index 00000000..bfcae6c5
--- /dev/null
+++ b/src/pulsecore/remap_mmx.c
@@ -0,0 +1,148 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk.com>
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2.1 of the License,
+  or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <string.h>
+
+#include <pulse/sample.h>
+#include <pulsecore/log.h>
+#include <pulsecore/macro.h>
+
+#include "cpu-x86.h"
+#include "remap.h"
+
+#define LOAD_SAMPLES                                   \
+                " movq (%1), %%mm0              \n\t"  \
+                " movq 8(%1), %%mm2             \n\t"  \
+                " movq 16(%1), %%mm4            \n\t"  \
+                " movq 24(%1), %%mm6            \n\t"  \
+                " movq %%mm0, %%mm1             \n\t"  \
+                " movq %%mm2, %%mm3             \n\t"  \
+                " movq %%mm4, %%mm5             \n\t"  \
+                " movq %%mm6, %%mm7             \n\t"
+
+#define UNPACK_SAMPLES(s)                              \
+                " punpckl"#s" %%mm0, %%mm0      \n\t"  \
+                " punpckh"#s" %%mm1, %%mm1      \n\t"  \
+                " punpckl"#s" %%mm2, %%mm2      \n\t"  \
+                " punpckh"#s" %%mm3, %%mm3      \n\t"  \
+                " punpckl"#s" %%mm4, %%mm4      \n\t"  \
+                " punpckh"#s" %%mm5, %%mm5      \n\t"  \
+                " punpckl"#s" %%mm6, %%mm6      \n\t"  \
+                " punpckh"#s" %%mm7, %%mm7      \n\t"  \
+
+#define STORE_SAMPLES                                  \
+                " movq %%mm0, (%0)              \n\t"  \
+                " movq %%mm1, 8(%0)             \n\t"  \
+                " movq %%mm2, 16(%0)            \n\t"  \
+                " movq %%mm3, 24(%0)            \n\t"  \
+                " movq %%mm4, 32(%0)            \n\t"  \
+                " movq %%mm5, 40(%0)            \n\t"  \
+                " movq %%mm6, 48(%0)            \n\t"  \
+                " movq %%mm7, 56(%0)            \n\t"  \
+                " add $32, %1                   \n\t"  \
+                " add $64, %0                   \n\t"
+
+#define HANDLE_SINGLE(s)                               \
+                " movd (%1), %%mm0              \n\t"  \
+                " movq %%mm0, %%mm1             \n\t"  \
+                " punpckl"#s" %%mm0, %%mm0      \n\t"  \
+                " movq %%mm0, (%0)              \n\t"  \
+                " add $4, %1                    \n\t"  \
+                " add $8, %0                    \n\t"
+
+#define MONO_TO_STEREO(s)                               \
+                " mov %3, %2                    \n\t"   \
+                " sar $3, %2                    \n\t"   \
+                " cmp $0, %2                    \n\t"   \
+                " je 2f                         \n\t"   \
+                "1:                             \n\t"   \
+                LOAD_SAMPLES                            \
+                UNPACK_SAMPLES(s)                       \
+                STORE_SAMPLES                           \
+                " dec %2                        \n\t"   \
+                " jne 1b                        \n\t"   \
+                "2:                             \n\t"   \
+                " mov %3, %2                    \n\t"   \
+                " and $7, %2                    \n\t"   \
+                " je 4f                         \n\t"   \
+                "3:                             \n\t"   \
+                HANDLE_SINGLE(s)                        \
+                " dec %2                        \n\t"   \
+                " jne 3b                        \n\t"   \
+                "4:                             \n\t"   \
+                " emms                          \n\t"
+
+static void remap_mono_to_stereo_mmx (pa_remap_t *m, void *dst, const void *src, unsigned n) {
+    pa_reg_x86 temp;
+
+    switch (*m->format) {
+        case PA_SAMPLE_FLOAT32NE:
+        {
+            __asm__ __volatile__ (
+                MONO_TO_STEREO(dq) /* do doubles to quads */
+                : "+r" (dst), "+r" (src), "=&r" (temp)
+                : "r" ((pa_reg_x86)n)
+                : "cc"
+            );
+            break;
+        }
+        case PA_SAMPLE_S16NE:
+        {
+            __asm__ __volatile__ (
+                MONO_TO_STEREO(wd) /* do words to doubles */
+                : "+r" (dst), "+r" (src), "=&r" (temp)
+                : "r" ((pa_reg_x86)n)
+                : "cc"
+            );
+            break;
+        }
+        default:
+            pa_assert_not_reached();
+    }
+}
+
+/* set the function that will execute the remapping based on the matrices */
+static void init_remap_mmx (pa_remap_t *m) {
+    unsigned n_oc, n_ic;
+
+    n_oc = m->o_ss->channels;
+    n_ic = m->i_ss->channels;
+
+    /* find some common channel remappings, fall back to full matrix operation. */
+    if (n_ic == 1 && n_oc == 2 &&
+            m->map_table_f[0][0] >= 1.0 && m->map_table_f[1][0] >= 1.0) {
+        m->do_remap = (pa_do_remap_func_t) remap_mono_to_stereo_mmx;
+        pa_log_info("Using MMX mono to stereo remapping");
+    }
+}
+
+void pa_remap_func_init_mmx (pa_cpu_x86_flag_t flags) {
+#if defined (__i386__) || defined (__amd64__)
+    pa_log_info("Initialising MMX optimized remappers.");
+
+    pa_set_init_remap_func ((pa_init_remap_func_t) init_remap_mmx);
+#endif /* defined (__i386__) || defined (__amd64__) */
+}
diff --git a/src/pulsecore/resampler.c b/src/pulsecore/resampler.c
index 59e0a0c1..f1bfa156 100644
--- a/src/pulsecore/resampler.c
+++ b/src/pulsecore/resampler.c
@@ -31,9 +31,6 @@
 
 #include <speex/speex_resampler.h>
 
-#include <liboil/liboilfuncs.h>
-#include <liboil/liboil.h>
-
 #include <pulse/xmalloc.h>
 #include <pulsecore/sconv.h>
 #include <pulsecore/log.h>
@@ -43,6 +40,7 @@
 #include "ffmpeg/avcodec.h"
 
 #include "resampler.h"
+#include "remap.h"
 
 /* Number of samples of extra space we allow the resamplers to return */
 #define EXTRA_FRAMES 128
@@ -64,7 +62,7 @@ struct pa_resampler {
     pa_convert_func_t to_work_format_func;
     pa_convert_func_t from_work_format_func;
 
-    float map_table[PA_CHANNELS_MAX][PA_CHANNELS_MAX];
+    pa_remap_t remap;
     pa_bool_t map_required;
 
     void (*impl_free)(pa_resampler *r);
@@ -214,6 +212,11 @@ pa_resampler* pa_resampler_new(
     r->i_ss = *a;
     r->o_ss = *b;
 
+    /* set up the remap structure */
+    r->remap.i_ss = &r->i_ss;
+    r->remap.o_ss = &r->o_ss;
+    r->remap.format = &r->work_format;
+
     if (am)
         r->i_cm = *am;
     else if (!pa_channel_map_init_auto(&r->i_cm, r->i_ss.channels, PA_CHANNEL_MAP_DEFAULT))
@@ -580,32 +583,41 @@ static int front_rear_side(pa_channel_position_t p) {
 
 static void calc_map_table(pa_resampler *r) {
     unsigned oc, ic;
+    unsigned n_oc, n_ic;
     pa_bool_t ic_connected[PA_CHANNELS_MAX];
     pa_bool_t remix;
     pa_strbuf *s;
     char *t;
+    pa_remap_t *m;
 
     pa_assert(r);
 
     if (!(r->map_required = (r->i_ss.channels != r->o_ss.channels || (!(r->flags & PA_RESAMPLER_NO_REMAP) && !pa_channel_map_equal(&r->i_cm, &r->o_cm)))))
         return;
 
-    memset(r->map_table, 0, sizeof(r->map_table));
+    m = &r->remap;
+
+    n_oc = r->o_ss.channels;
+    n_ic = r->i_ss.channels;
+
+    memset(m->map_table_f, 0, sizeof(m->map_table_f));
+    memset(m->map_table_i, 0, sizeof(m->map_table_i));
+
     memset(ic_connected, 0, sizeof(ic_connected));
     remix = (r->flags & (PA_RESAMPLER_NO_REMAP|PA_RESAMPLER_NO_REMIX)) == 0;
 
-    for (oc = 0; oc < r->o_ss.channels; oc++) {
+    for (oc = 0; oc < n_oc; oc++) {
         pa_bool_t oc_connected = FALSE;
         pa_channel_position_t b = r->o_cm.map[oc];
 
-        for (ic = 0; ic < r->i_ss.channels; ic++) {
+        for (ic = 0; ic < n_ic; ic++) {
             pa_channel_position_t a = r->i_cm.map[ic];
 
             if (r->flags & PA_RESAMPLER_NO_REMAP) {
                 /* We shall not do any remapping. Hence, just check by index */
 
                 if (ic == oc)
-                    r->map_table[oc][ic] = 1.0;
+                    m->map_table_f[oc][ic] = 1.0;
 
                 continue;
             }
@@ -614,7 +626,7 @@ static void calc_map_table(pa_resampler *r) {
                 /* We shall not do any remixing. Hence, just check by name */
 
                 if (a == b)
-                    r->map_table[oc][ic] = 1.0;
+                    m->map_table_f[oc][ic] = 1.0;
 
                 continue;
             }
@@ -689,7 +701,7 @@ static void calc_map_table(pa_resampler *r) {
              */
 
             if (a == b || a == PA_CHANNEL_POSITION_MONO || b == PA_CHANNEL_POSITION_MONO) {
-                r->map_table[oc][ic] = 1.0;
+                m->map_table_f[oc][ic] = 1.0;
 
                 oc_connected = TRUE;
                 ic_connected[ic] = TRUE;
@@ -707,14 +719,14 @@ static void calc_map_table(pa_resampler *r) {
                 /* We are not connected and on the left side, let's
                  * average all left side input channels. */
 
-                for (ic = 0; ic < r->i_ss.channels; ic++)
+                for (ic = 0; ic < n_ic; ic++)
                     if (on_left(r->i_cm.map[ic]))
                         n++;
 
                 if (n > 0)
-                    for (ic = 0; ic < r->i_ss.channels; ic++)
+                    for (ic = 0; ic < n_ic; ic++)
                         if (on_left(r->i_cm.map[ic])) {
-                            r->map_table[oc][ic] = 1.0f / (float) n;
+                            m->map_table_f[oc][ic] = 1.0f / (float) n;
                             ic_connected[ic] = TRUE;
                         }
 
@@ -728,14 +740,14 @@ static void calc_map_table(pa_resampler *r) {
                 /* We are not connected and on the right side, let's
                  * average all right side input channels. */
 
-                for (ic = 0; ic < r->i_ss.channels; ic++)
+                for (ic = 0; ic < n_ic; ic++)
                     if (on_right(r->i_cm.map[ic]))
                         n++;
 
                 if (n > 0)
-                    for (ic = 0; ic < r->i_ss.channels; ic++)
+                    for (ic = 0; ic < n_ic; ic++)
                         if (on_right(r->i_cm.map[ic])) {
-                            r->map_table[oc][ic] = 1.0f / (float) n;
+                            m->map_table_f[oc][ic] = 1.0f / (float) n;
                             ic_connected[ic] = TRUE;
                         }
 
@@ -749,14 +761,14 @@ static void calc_map_table(pa_resampler *r) {
                 /* We are not connected and at the center. Let's
                  * average all center input channels. */
 
-                for (ic = 0; ic < r->i_ss.channels; ic++)
+                for (ic = 0; ic < n_ic; ic++)
                     if (on_center(r->i_cm.map[ic]))
                         n++;
 
                 if (n > 0) {
-                    for (ic = 0; ic < r->i_ss.channels; ic++)
+                    for (ic = 0; ic < n_ic; ic++)
                         if (on_center(r->i_cm.map[ic])) {
-                            r->map_table[oc][ic] = 1.0f / (float) n;
+                            m->map_table_f[oc][ic] = 1.0f / (float) n;
                             ic_connected[ic] = TRUE;
                         }
                 } else {
@@ -766,14 +778,14 @@ static void calc_map_table(pa_resampler *r) {
 
                     n = 0;
 
-                    for (ic = 0; ic < r->i_ss.channels; ic++)
+                    for (ic = 0; ic < n_ic; ic++)
                         if (on_left(r->i_cm.map[ic]) || on_right(r->i_cm.map[ic]))
                             n++;
 
                     if (n > 0)
-                        for (ic = 0; ic < r->i_ss.channels; ic++)
+                        for (ic = 0; ic < n_ic; ic++)
                             if (on_left(r->i_cm.map[ic]) || on_right(r->i_cm.map[ic])) {
-                                r->map_table[oc][ic] = 1.0f / (float) n;
+                                m->map_table_f[oc][ic] = 1.0f / (float) n;
                                 ic_connected[ic] = TRUE;
                             }
 
@@ -787,12 +799,12 @@ static void calc_map_table(pa_resampler *r) {
                 /* We are not connected and an LFE. Let's average all
                  * channels for LFE. */
 
-                for (ic = 0; ic < r->i_ss.channels; ic++) {
+                for (ic = 0; ic < n_ic; ic++) {
 
                     if (!(r->flags & PA_RESAMPLER_NO_LFE))
-                        r->map_table[oc][ic] = 1.0f / (float) r->i_ss.channels;
+                        m->map_table_f[oc][ic] = 1.0f / (float) n_ic;
                     else
-                        r->map_table[oc][ic] = 0;
+                        m->map_table_f[oc][ic] = 0;
 
                     /* Please note that a channel connected to LFE
                      * doesn't really count as connected. */
@@ -808,7 +820,7 @@ static void calc_map_table(pa_resampler *r) {
             ic_unconnected_center = 0,
             ic_unconnected_lfe = 0;
 
-        for (ic = 0; ic < r->i_ss.channels; ic++) {
+        for (ic = 0; ic < n_ic; ic++) {
             pa_channel_position_t a = r->i_cm.map[ic];
 
             if (ic_connected[ic])
@@ -831,20 +843,20 @@ static void calc_map_table(pa_resampler *r) {
              * the left side by .9 and add in our averaged unconnected
              * channels multplied by .1 */
 
-            for (oc = 0; oc < r->o_ss.channels; oc++) {
+            for (oc = 0; oc < n_oc; oc++) {
 
                 if (!on_left(r->o_cm.map[oc]))
                     continue;
 
-                for (ic = 0; ic < r->i_ss.channels; ic++) {
+                for (ic = 0; ic < n_ic; ic++) {
 
                     if (ic_connected[ic]) {
-                        r->map_table[oc][ic] *= .9f;
+                        m->map_table_f[oc][ic] *= .9f;
                         continue;
                     }
 
                     if (on_left(r->i_cm.map[ic]))
-                        r->map_table[oc][ic] = .1f / (float) ic_unconnected_left;
+                        m->map_table_f[oc][ic] = .1f / (float) ic_unconnected_left;
                 }
             }
         }
@@ -856,20 +868,20 @@ static void calc_map_table(pa_resampler *r) {
              * the right side by .9 and add in our averaged unconnected
              * channels multplied by .1 */
 
-            for (oc = 0; oc < r->o_ss.channels; oc++) {
+            for (oc = 0; oc < n_oc; oc++) {
 
                 if (!on_right(r->o_cm.map[oc]))
                     continue;
 
-                for (ic = 0; ic < r->i_ss.channels; ic++) {
+                for (ic = 0; ic < n_ic; ic++) {
 
                     if (ic_connected[ic]) {
-                        r->map_table[oc][ic] *= .9f;
+                        m->map_table_f[oc][ic] *= .9f;
                         continue;
                     }
 
                     if (on_right(r->i_cm.map[ic]))
-                        r->map_table[oc][ic] = .1f / (float) ic_unconnected_right;
+                        m->map_table_f[oc][ic] = .1f / (float) ic_unconnected_right;
                 }
             }
         }
@@ -882,20 +894,20 @@ static void calc_map_table(pa_resampler *r) {
              * the center side by .9 and add in our averaged unconnected
              * channels multplied by .1 */
 
-            for (oc = 0; oc < r->o_ss.channels; oc++) {
+            for (oc = 0; oc < n_oc; oc++) {
 
                 if (!on_center(r->o_cm.map[oc]))
                     continue;
 
-                for (ic = 0; ic < r->i_ss.channels; ic++)  {
+                for (ic = 0; ic < n_ic; ic++)  {
 
                     if (ic_connected[ic]) {
-                        r->map_table[oc][ic] *= .9f;
+                        m->map_table_f[oc][ic] *= .9f;
                         continue;
                     }
 
                     if (on_center(r->i_cm.map[ic])) {
-                        r->map_table[oc][ic] = .1f / (float) ic_unconnected_center;
+                        m->map_table_f[oc][ic] = .1f / (float) ic_unconnected_center;
                         mixed_in = TRUE;
                     }
                 }
@@ -913,7 +925,7 @@ static void calc_map_table(pa_resampler *r) {
                    it into left and right. Using .375 and 0.75 as
                    factors. */
 
-                for (ic = 0; ic < r->i_ss.channels; ic++) {
+                for (ic = 0; ic < n_ic; ic++) {
 
                     if (ic_connected[ic])
                         continue;
@@ -921,7 +933,7 @@ static void calc_map_table(pa_resampler *r) {
                     if (!on_center(r->i_cm.map[ic]))
                         continue;
 
-                    for (oc = 0; oc < r->o_ss.channels; oc++) {
+                    for (oc = 0; oc < n_oc; oc++) {
 
                         if (!on_left(r->o_cm.map[oc]) && !on_right(r->o_cm.map[oc]))
                             continue;
@@ -932,7 +944,7 @@ static void calc_map_table(pa_resampler *r) {
                         }
                     }
 
-                    for (oc = 0; oc < r->o_ss.channels; oc++) {
+                    for (oc = 0; oc < n_oc; oc++) {
 
                         if (!on_left(r->o_cm.map[oc]) && !on_right(r->o_cm.map[oc]))
                             continue;
@@ -942,7 +954,7 @@ static void calc_map_table(pa_resampler *r) {
                     }
                 }
 
-                for (oc = 0; oc < r->o_ss.channels; oc++) {
+                for (oc = 0; oc < n_oc; oc++) {
 
                     if (!on_left(r->o_cm.map[oc]) && !on_right(r->o_cm.map[oc]))
                         continue;
@@ -950,10 +962,10 @@ static void calc_map_table(pa_resampler *r) {
                     if (ncenter[oc] <= 0)
                         continue;
 
-                    for (ic = 0; ic < r->i_ss.channels; ic++)  {
+                    for (ic = 0; ic < n_ic; ic++)  {
 
                         if (ic_connected[ic]) {
-                            r->map_table[oc][ic] *= .75f;
+                            m->map_table_f[oc][ic] *= .75f;
                             continue;
                         }
 
@@ -961,7 +973,7 @@ static void calc_map_table(pa_resampler *r) {
                             continue;
 
                         if (!found_frs[ic] || front_rear_side(r->i_cm.map[ic]) == front_rear_side(r->o_cm.map[oc]))
-                            r->map_table[oc][ic] = .375f / (float) ncenter[oc];
+                            m->map_table_f[oc][ic] = .375f / (float) ncenter[oc];
                     }
                 }
             }
@@ -972,40 +984,46 @@ static void calc_map_table(pa_resampler *r) {
             /* OK, so there is an unconnected LFE channel. Let's mix
              * it into all channels, with factor 0.375 */
 
-            for (ic = 0; ic < r->i_ss.channels; ic++)  {
+            for (ic = 0; ic < n_ic; ic++)  {
 
                 if (!on_lfe(r->i_cm.map[ic]))
                     continue;
 
-                for (oc = 0; oc < r->o_ss.channels; oc++)
-                    r->map_table[oc][ic] = 0.375f / (float) ic_unconnected_lfe;
+                for (oc = 0; oc < n_oc; oc++)
+                    m->map_table_f[oc][ic] = 0.375f / (float) ic_unconnected_lfe;
             }
         }
     }
-
+    /* make an 16:16 int version of the matrix */
+    for (oc = 0; oc < n_oc; oc++)
+        for (ic = 0; ic < n_ic; ic++)
+            m->map_table_i[oc][ic] = (int32_t) (m->map_table_f[oc][ic] * 0x10000);
 
     s = pa_strbuf_new();
 
     pa_strbuf_printf(s, "     ");
-    for (ic = 0; ic < r->i_ss.channels; ic++)
+    for (ic = 0; ic < n_ic; ic++)
         pa_strbuf_printf(s, "  I%02u ", ic);
     pa_strbuf_puts(s, "\n    +");
 
-    for (ic = 0; ic < r->i_ss.channels; ic++)
+    for (ic = 0; ic < n_ic; ic++)
         pa_strbuf_printf(s, "------");
     pa_strbuf_puts(s, "\n");
 
-    for (oc = 0; oc < r->o_ss.channels; oc++) {
+    for (oc = 0; oc < n_oc; oc++) {
         pa_strbuf_printf(s, "O%02u |", oc);
 
-        for (ic = 0; ic < r->i_ss.channels; ic++)
-            pa_strbuf_printf(s, " %1.3f", r->map_table[oc][ic]);
+        for (ic = 0; ic < n_ic; ic++)
+            pa_strbuf_printf(s, " %1.3f", m->map_table_f[oc][ic]);
 
         pa_strbuf_puts(s, "\n");
     }
 
     pa_log_debug("Channel matrix:\n%s", t = pa_strbuf_tostring_free(s));
     pa_xfree(t);
+
+    /* initialize the remapping function */
+    pa_init_remap (m);
 }
 
 static pa_memchunk* convert_to_work_format(pa_resampler *r, pa_memchunk *input) {
@@ -1045,41 +1063,10 @@ static pa_memchunk* convert_to_work_format(pa_resampler *r, pa_memchunk *input)
     return &r->buf1;
 }
 
-static void vectoradd_s16_with_fraction(
-        int16_t *d, int dstr,
-        const int16_t *s1, int sstr1,
-        const int16_t *s2, int sstr2,
-        int n,
-        float s3, float s4) {
-
-    int32_t i3, i4;
-
-    i3 = (int32_t) (s3 * 0x10000);
-    i4 = (int32_t) (s4 * 0x10000);
-
-    for (; n > 0; n--) {
-        int32_t a, b;
-
-        a = *s1;
-        b = *s2;
-
-        a = (a * i3) / 0x10000;
-        b = (b * i4) / 0x10000;
-
-        *d = (int16_t) (a + b);
-
-        s1 = (const int16_t*) ((const uint8_t*) s1 + sstr1);
-        s2 = (const int16_t*) ((const uint8_t*) s2 + sstr2);
-        d = (int16_t*) ((uint8_t*) d + dstr);
-
-    }
-}
-
 static pa_memchunk *remap_channels(pa_resampler *r, pa_memchunk *input) {
     unsigned in_n_samples, out_n_samples, n_frames;
-    int i_skip, o_skip;
-    unsigned oc;
     void *src, *dst;
+    pa_remap_t *remap;
 
     pa_assert(r);
     pa_assert(input);
@@ -1108,76 +1095,14 @@ static pa_memchunk *remap_channels(pa_resampler *r, pa_memchunk *input) {
     src = ((uint8_t*) pa_memblock_acquire(input->memblock) + input->index);
     dst = pa_memblock_acquire(r->buf2.memblock);
 
-    memset(dst, 0, r->buf2.length);
-
-    o_skip = (int) (r->w_sz * r->o_ss.channels);
-    i_skip = (int) (r->w_sz * r->i_ss.channels);
-
-    switch (r->work_format) {
-        case PA_SAMPLE_FLOAT32NE:
-
-            for (oc = 0; oc < r->o_ss.channels; oc++) {
-                unsigned ic;
-                static const float one = 1.0;
-
-                for (ic = 0; ic < r->i_ss.channels; ic++) {
+    remap = &r->remap;
 
-                    if (r->map_table[oc][ic] <= 0.0)
-                        continue;
-
-                    oil_vectoradd_f32(
-                            (float*) dst + oc, o_skip,
-                            (float*) dst + oc, o_skip,
-                            (float*) src + ic, i_skip,
-                            (int) n_frames,
-                            &one, &r->map_table[oc][ic]);
-                }
-            }
-
-            break;
-
-        case PA_SAMPLE_S16NE:
-
-            for (oc = 0; oc < r->o_ss.channels; oc++) {
-                unsigned ic;
-
-                for (ic = 0; ic < r->i_ss.channels; ic++) {
-
-                    if (r->map_table[oc][ic] <= 0.0)
-                        continue;
-
-                    if (r->map_table[oc][ic] >= 1.0) {
-                        static const int16_t one = 1;
-
-                        oil_vectoradd_s16(
-                                (int16_t*) dst + oc, o_skip,
-                                (int16_t*) dst + oc, o_skip,
-                                (int16_t*) src + ic, i_skip,
-                                (int) n_frames,
-                                &one, &one);
-
-                    } else
-
-                        vectoradd_s16_with_fraction(
-                                (int16_t*) dst + oc, o_skip,
-                                (int16_t*) dst + oc, o_skip,
-                                (int16_t*) src + ic, i_skip,
-                                (int) n_frames,
-                                1.0f, r->map_table[oc][ic]);
-                }
-            }
-
-            break;
-
-        default:
-            pa_assert_not_reached();
-    }
+    pa_assert (remap->do_remap);
+    remap->do_remap (remap, dst, src, n_frames);
 
     pa_memblock_release(input->memblock);
     pa_memblock_release(r->buf2.memblock);
 
-    r->buf2.length = out_n_samples * r->w_sz;
-
     return &r->buf2;
 }
 
@@ -1469,7 +1394,7 @@ static void trivial_resample(pa_resampler *r, const pa_memchunk *input, unsigned
 
         pa_assert(o_index * fz < pa_memblock_get_length(output->memblock));
 
-        oil_memcpy((uint8_t*) dst + fz * o_index,
+        memcpy((uint8_t*) dst + fz * o_index,
                    (uint8_t*) src + fz * j, (int) fz);
     }
 
diff --git a/src/pulsecore/rtpoll.c b/src/pulsecore/rtpoll.c
index 42708a8a..666cbc98 100644
--- a/src/pulsecore/rtpoll.c
+++ b/src/pulsecore/rtpoll.c
@@ -63,6 +63,7 @@ struct pa_rtpoll {
     pa_bool_t running:1;
     pa_bool_t rebuild_needed:1;
     pa_bool_t quit:1;
+    pa_bool_t timer_elapsed:1;
 
 #ifdef DEBUG_TIMING
     pa_usec_t timestamp;
@@ -94,26 +95,14 @@ PA_STATIC_FLIST_DECLARE(items, 0, pa_xfree);
 pa_rtpoll *pa_rtpoll_new(void) {
     pa_rtpoll *p;
 
-    p = pa_xnew(pa_rtpoll, 1);
+    p = pa_xnew0(pa_rtpoll, 1);
 
     p->n_pollfd_alloc = 32;
     p->pollfd = pa_xnew(struct pollfd, p->n_pollfd_alloc);
     p->pollfd2 = pa_xnew(struct pollfd, p->n_pollfd_alloc);
-    p->n_pollfd_used = 0;
-
-    pa_zero(p->next_elapse);
-    p->timer_enabled = FALSE;
-
-    p->running = FALSE;
-    p->scan_for_dead = FALSE;
-    p->rebuild_needed = FALSE;
-    p->quit = FALSE;
-
-    PA_LLIST_HEAD_INIT(pa_rtpoll_item, p->items);
 
 #ifdef DEBUG_TIMING
     p->timestamp = pa_rtclock_now();
-    p->slept = p->awake = 0;
 #endif
 
     return p;
@@ -229,6 +218,7 @@ int pa_rtpoll_run(pa_rtpoll *p, pa_bool_t wait_op) {
     pa_assert(!p->running);
 
     p->running = TRUE;
+    p->timer_elapsed = FALSE;
 
     /* First, let's do some work */
     for (i = p->items; i && i->priority < PA_RTPOLL_NEVER; i = i->next) {
@@ -286,7 +276,7 @@ int pa_rtpoll_run(pa_rtpoll *p, pa_bool_t wait_op) {
     if (p->rebuild_needed)
         rtpoll_rebuild(p);
 
-    memset(&timeout, 0, sizeof(timeout));
+    pa_zero(timeout);
 
     /* Calculate timeout */
     if (wait_op && !p->quit && p->timer_enabled) {
@@ -314,9 +304,11 @@ int pa_rtpoll_run(pa_rtpoll *p, pa_bool_t wait_op) {
         r = ppoll(p->pollfd, p->n_pollfd_used, (!wait_op || p->quit || p->timer_enabled) ? &ts : NULL, NULL);
     }
 #else
-        r = poll(p->pollfd, p->n_pollfd_used, (!wait_op || p->quit || p->timer_enabled) ? (int) ((timeout.tv_sec*1000) + (timeout.tv_usec / 1000)) : -1);
+    r = poll(p->pollfd, p->n_pollfd_used, (!wait_op || p->quit || p->timer_enabled) ? (int) ((timeout.tv_sec*1000) + (timeout.tv_usec / 1000)) : -1);
 #endif
 
+    p->timer_elapsed = r == 0;
+
 #ifdef DEBUG_TIMING
     {
         pa_usec_t now = pa_rtclock_now();
@@ -628,3 +620,9 @@ void pa_rtpoll_quit(pa_rtpoll *p) {
 
     p->quit = TRUE;
 }
+
+pa_bool_t pa_rtpoll_timer_elapsed(pa_rtpoll *p) {
+    pa_assert(p);
+
+    return p->timer_elapsed;
+}
diff --git a/src/pulsecore/rtpoll.h b/src/pulsecore/rtpoll.h
index d2d69cad..b2a87fca 100644
--- a/src/pulsecore/rtpoll.h
+++ b/src/pulsecore/rtpoll.h
@@ -73,6 +73,10 @@ void pa_rtpoll_set_timer_absolute(pa_rtpoll *p, pa_usec_t usec);
 void pa_rtpoll_set_timer_relative(pa_rtpoll *p, pa_usec_t usec);
 void pa_rtpoll_set_timer_disabled(pa_rtpoll *p);
 
+/* Return TRUE when the elapsed timer was the reason for
+ * the last pa_rtpoll_run() invocation to finish */
+pa_bool_t pa_rtpoll_timer_elapsed(pa_rtpoll *p);
+
 /* A new fd wakeup item for pa_rtpoll */
 pa_rtpoll_item *pa_rtpoll_item_new(pa_rtpoll *p, pa_rtpoll_priority_t prio, unsigned n_fds);
 void pa_rtpoll_item_free(pa_rtpoll_item *i);
diff --git a/src/pulsecore/sample-util.c b/src/pulsecore/sample-util.c
index 5b8ccf59..6e97e5a9 100644
--- a/src/pulsecore/sample-util.c
+++ b/src/pulsecore/sample-util.c
@@ -30,9 +30,6 @@
 #include <stdio.h>
 #include <errno.h>
 
-#include <liboil/liboilfuncs.h>
-#include <liboil/liboil.h>
-
 #include <pulse/timeval.h>
 
 #include <pulsecore/log.h>
@@ -106,24 +103,36 @@ void* pa_silence_memory(void *p, size_t length, const pa_sample_spec *spec) {
     return p;
 }
 
+#define VOLUME_PADDING 32
+
 static void calc_linear_integer_volume(int32_t linear[], const pa_cvolume *volume) {
-    unsigned channel;
+    unsigned channel, nchannels, padding;
 
     pa_assert(linear);
     pa_assert(volume);
 
-    for (channel = 0; channel < volume->channels; channel++)
+    nchannels = volume->channels;
+
+    for (channel = 0; channel < nchannels; channel++)
         linear[channel] = (int32_t) lrint(pa_sw_volume_to_linear(volume->values[channel]) * 0x10000);
+
+    for (padding = 0; padding < VOLUME_PADDING; padding++, channel++)
+        linear[channel] = linear[padding];
 }
 
 static void calc_linear_float_volume(float linear[], const pa_cvolume *volume) {
-    unsigned channel;
+    unsigned channel, nchannels, padding;
 
     pa_assert(linear);
     pa_assert(volume);
 
-    for (channel = 0; channel < volume->channels; channel++)
+    nchannels = volume->channels;
+
+    for (channel = 0; channel < nchannels; channel++)
         linear[channel] = (float) pa_sw_volume_to_linear(volume->values[channel]);
+
+    for (padding = 0; padding < VOLUME_PADDING; padding++, channel++)
+        linear[channel] = linear[padding];
 }
 
 static void calc_linear_integer_stream_volumes(pa_mix_info streams[], unsigned nstreams, const pa_cvolume *volume, const pa_sample_spec *spec) {
@@ -690,6 +699,28 @@ size_t pa_mix(
     return length;
 }
 
+typedef union {
+  float f;
+  uint32_t i;
+} volume_val;
+
+typedef void (*pa_calc_volume_func_t) (void *volumes, const pa_cvolume *volume);
+
+static const pa_calc_volume_func_t calc_volume_table[] = {
+  [PA_SAMPLE_U8]        = (pa_calc_volume_func_t) calc_linear_integer_volume,
+  [PA_SAMPLE_ALAW]      = (pa_calc_volume_func_t) calc_linear_integer_volume,
+  [PA_SAMPLE_ULAW]      = (pa_calc_volume_func_t) calc_linear_integer_volume,
+  [PA_SAMPLE_S16LE]     = (pa_calc_volume_func_t) calc_linear_integer_volume,
+  [PA_SAMPLE_S16BE]     = (pa_calc_volume_func_t) calc_linear_integer_volume,
+  [PA_SAMPLE_FLOAT32LE] = (pa_calc_volume_func_t) calc_linear_float_volume,
+  [PA_SAMPLE_FLOAT32BE] = (pa_calc_volume_func_t) calc_linear_float_volume,
+  [PA_SAMPLE_S32LE]     = (pa_calc_volume_func_t) calc_linear_integer_volume,
+  [PA_SAMPLE_S32BE]     = (pa_calc_volume_func_t) calc_linear_integer_volume,
+  [PA_SAMPLE_S24LE]     = (pa_calc_volume_func_t) calc_linear_integer_volume,
+  [PA_SAMPLE_S24BE]     = (pa_calc_volume_func_t) calc_linear_integer_volume,
+  [PA_SAMPLE_S24_32LE]  = (pa_calc_volume_func_t) calc_linear_integer_volume,
+  [PA_SAMPLE_S24_32BE]  = (pa_calc_volume_func_t) calc_linear_integer_volume
+};
 
 void pa_volume_memchunk(
         pa_memchunk*c,
@@ -697,6 +728,8 @@ void pa_volume_memchunk(
         const pa_cvolume *volume) {
 
     void *ptr;
+    volume_val linear[PA_CHANNELS_MAX + VOLUME_PADDING];
+    pa_do_volume_func_t do_volume;
 
     pa_assert(c);
     pa_assert(spec);
@@ -714,337 +747,19 @@ void pa_volume_memchunk(
         return;
     }
 
-    ptr = (uint8_t*) pa_memblock_acquire(c->memblock) + c->index;
-
-    switch (spec->format) {
-
-        case PA_SAMPLE_S16NE: {
-            int16_t *d, *e;
-            unsigned channel;
-            int32_t linear[PA_CHANNELS_MAX];
-
-            calc_linear_integer_volume(linear, volume);
-
-            e = (int16_t*) ptr + c->length/sizeof(int16_t);
-
-            for (channel = 0, d = ptr; d < e; d++) {
-                int32_t t, hi, lo;
-
-                /* Multiplying the 32bit volume factor with the 16bit
-                 * sample might result in an 48bit value. We want to
-                 * do without 64 bit integers and hence do the
-                 * multiplication independantly for the HI and LO part
-                 * of the volume. */
-
-                hi = linear[channel] >> 16;
-                lo = linear[channel] & 0xFFFF;
-
-                t = (int32_t)(*d);
-                t = ((t * lo) >> 16) + (t * hi);
-                t = PA_CLAMP_UNLIKELY(t, -0x8000, 0x7FFF);
-                *d = (int16_t) t;
-
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-
-            break;
-        }
-
-        case PA_SAMPLE_S16RE: {
-            int16_t *d, *e;
-            unsigned channel;
-            int32_t linear[PA_CHANNELS_MAX];
-
-            calc_linear_integer_volume(linear, volume);
-
-            e = (int16_t*) ptr + c->length/sizeof(int16_t);
-
-            for (channel = 0, d = ptr; d < e; d++) {
-                int32_t t, hi, lo;
-
-                hi = linear[channel] >> 16;
-                lo = linear[channel] & 0xFFFF;
-
-                t = (int32_t) PA_INT16_SWAP(*d);
-                t = ((t * lo) >> 16) + (t * hi);
-                t = PA_CLAMP_UNLIKELY(t, -0x8000, 0x7FFF);
-                *d = PA_INT16_SWAP((int16_t) t);
-
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-
-            break;
-        }
-
-        case PA_SAMPLE_S32NE: {
-            int32_t *d, *e;
-            unsigned channel;
-            int32_t linear[PA_CHANNELS_MAX];
-
-            calc_linear_integer_volume(linear, volume);
-
-            e = (int32_t*) ptr + c->length/sizeof(int32_t);
-
-            for (channel = 0, d = ptr; d < e; d++) {
-                int64_t t;
-
-                t = (int64_t)(*d);
-                t = (t * linear[channel]) >> 16;
-                t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
-                *d = (int32_t) t;
-
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-            break;
-        }
-
-        case PA_SAMPLE_S32RE: {
-            int32_t *d, *e;
-            unsigned channel;
-            int32_t linear[PA_CHANNELS_MAX];
-
-            calc_linear_integer_volume(linear, volume);
-
-            e = (int32_t*) ptr + c->length/sizeof(int32_t);
-
-            for (channel = 0, d = ptr; d < e; d++) {
-                int64_t t;
-
-                t = (int64_t) PA_INT32_SWAP(*d);
-                t = (t * linear[channel]) >> 16;
-                t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
-                *d = PA_INT32_SWAP((int32_t) t);
-
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-            break;
-        }
-
-        case PA_SAMPLE_S24NE: {
-            uint8_t *d, *e;
-            unsigned channel;
-            int32_t linear[PA_CHANNELS_MAX];
-
-            calc_linear_integer_volume(linear, volume);
-
-            e = (uint8_t*) ptr + c->length;
-
-            for (channel = 0, d = ptr; d < e; d += 3) {
-                int64_t t;
-
-                t = (int64_t)((int32_t) (PA_READ24NE(d) << 8));
-                t = (t * linear[channel]) >> 16;
-                t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
-                PA_WRITE24NE(d, ((uint32_t) (int32_t) t) >> 8);
-
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-            break;
-        }
-
-        case PA_SAMPLE_S24RE: {
-            uint8_t *d, *e;
-            unsigned channel;
-            int32_t linear[PA_CHANNELS_MAX];
-
-            calc_linear_integer_volume(linear, volume);
-
-            e = (uint8_t*) ptr + c->length;
-
-            for (channel = 0, d = ptr; d < e; d += 3) {
-                int64_t t;
-
-                t = (int64_t)((int32_t) (PA_READ24RE(d) << 8));
-                t = (t * linear[channel]) >> 16;
-                t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
-                PA_WRITE24RE(d, ((uint32_t) (int32_t) t) >> 8);
-
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-            break;
-        }
-
-        case PA_SAMPLE_S24_32NE: {
-            uint32_t *d, *e;
-            unsigned channel;
-            int32_t linear[PA_CHANNELS_MAX];
-
-            calc_linear_integer_volume(linear, volume);
-
-            e = (uint32_t*) ptr + c->length/sizeof(uint32_t);
-
-            for (channel = 0, d = ptr; d < e; d++) {
-                int64_t t;
-
-                t = (int64_t) ((int32_t) (*d << 8));
-                t = (t * linear[channel]) >> 16;
-                t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
-                *d = ((uint32_t) ((int32_t) t)) >> 8;
-
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-            break;
-        }
-
-        case PA_SAMPLE_S24_32RE: {
-            uint32_t *d, *e;
-            unsigned channel;
-            int32_t linear[PA_CHANNELS_MAX];
-
-            calc_linear_integer_volume(linear, volume);
-
-            e = (uint32_t*) ptr + c->length/sizeof(uint32_t);
-
-            for (channel = 0, d = ptr; d < e; d++) {
-                int64_t t;
-
-                t = (int64_t) ((int32_t) (PA_UINT32_SWAP(*d) << 8));
-                t = (t * linear[channel]) >> 16;
-                t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
-                *d = PA_UINT32_SWAP(((uint32_t) ((int32_t) t)) >> 8);
-
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-            break;
-        }
-
-        case PA_SAMPLE_U8: {
-            uint8_t *d, *e;
-            unsigned channel;
-            int32_t linear[PA_CHANNELS_MAX];
-
-            calc_linear_integer_volume(linear, volume);
-
-            e = (uint8_t*) ptr + c->length;
-
-            for (channel = 0, d = ptr; d < e; d++) {
-                int32_t t, hi, lo;
-
-                hi = linear[channel] >> 16;
-                lo = linear[channel] & 0xFFFF;
-
-                t = (int32_t) *d - 0x80;
-                t = ((t * lo) >> 16) + (t * hi);
-                t = PA_CLAMP_UNLIKELY(t, -0x80, 0x7F);
-                *d = (uint8_t) (t + 0x80);
-
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-            break;
-        }
-
-        case PA_SAMPLE_ULAW: {
-            uint8_t *d, *e;
-            unsigned channel;
-            int32_t linear[PA_CHANNELS_MAX];
-
-            calc_linear_integer_volume(linear, volume);
-
-            e = (uint8_t*) ptr + c->length;
-
-            for (channel = 0, d = ptr; d < e; d++) {
-                int32_t t, hi, lo;
-
-                hi = linear[channel] >> 16;
-                lo = linear[channel] & 0xFFFF;
-
-                t = (int32_t) st_ulaw2linear16(*d);
-                t = ((t * lo) >> 16) + (t * hi);
-                t = PA_CLAMP_UNLIKELY(t, -0x8000, 0x7FFF);
-                *d = (uint8_t) st_14linear2ulaw((int16_t) t >> 2);
-
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-            break;
-        }
-
-        case PA_SAMPLE_ALAW: {
-            uint8_t *d, *e;
-            unsigned channel;
-            int32_t linear[PA_CHANNELS_MAX];
-
-            calc_linear_integer_volume(linear, volume);
-
-            e = (uint8_t*) ptr + c->length;
-
-            for (channel = 0, d = ptr; d < e; d++) {
-                int32_t t, hi, lo;
-
-                hi = linear[channel] >> 16;
-                lo = linear[channel] & 0xFFFF;
-
-                t = (int32_t) st_alaw2linear16(*d);
-                t = ((t * lo) >> 16) + (t * hi);
-                t = PA_CLAMP_UNLIKELY(t, -0x8000, 0x7FFF);
-                *d = (uint8_t) st_13linear2alaw((int16_t) t >> 3);
-
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-            break;
-        }
-
-        case PA_SAMPLE_FLOAT32NE: {
-            float *d;
-            int skip;
-            unsigned n;
-            unsigned channel;
-
-            d = ptr;
-            skip = (int) (spec->channels * sizeof(float));
-            n = (unsigned) (c->length/sizeof(float)/spec->channels);
-
-            for (channel = 0; channel < spec->channels; channel ++) {
-                float v, *t;
-
-                if (PA_UNLIKELY(volume->values[channel] == PA_VOLUME_NORM))
-                    continue;
-
-                v = (float) pa_sw_volume_to_linear(volume->values[channel]);
-                t = d + channel;
-                oil_scalarmult_f32(t, skip, t, skip, &v, (int) n);
-            }
-            break;
-        }
-
-        case PA_SAMPLE_FLOAT32RE: {
-            float *d, *e;
-            unsigned channel;
-            float linear[PA_CHANNELS_MAX];
-
-            calc_linear_float_volume(linear, volume);
-
-            e = (float*) ptr + c->length/sizeof(float);
-
-            for (channel = 0, d = ptr; d < e; d++) {
-                float t;
+    if (spec->format < 0 || spec->format > PA_SAMPLE_MAX) {
+      pa_log_warn(" Unable to change volume of format %s.", pa_sample_format_to_string(spec->format));
+      return;
+    }
 
-                t = PA_FLOAT32_SWAP(*d);
-                t *= linear[channel];
-                *d = PA_FLOAT32_SWAP(t);
+    do_volume = pa_get_volume_func (spec->format);
+    pa_assert(do_volume);
 
-                if (PA_UNLIKELY(++channel >= spec->channels))
-                    channel = 0;
-            }
-
-            break;
-        }
+    calc_volume_table[spec->format] ((void *)linear, volume);
 
+    ptr = (uint8_t*) pa_memblock_acquire(c->memblock) + c->index;
 
-        default:
-            pa_log_warn(" Unable to change volume of format %s.", pa_sample_format_to_string(spec->format));
-            /* If we cannot change the volume, we just don't do it */
-    }
+    do_volume (ptr, (void *)linear, spec->channels, c->length);
 
     pa_memblock_release(c->memblock);
 }
@@ -1090,7 +805,7 @@ void pa_interleave(const void *src[], unsigned channels, void *dst, size_t ss, u
         d = (uint8_t*) dst + c * ss;
 
         for (j = 0; j < n; j ++) {
-            oil_memcpy(d, s, (int) ss);
+            memcpy(d, s, (int) ss);
             s = (uint8_t*) s + ss;
             d = (uint8_t*) d + fs;
         }
@@ -1118,7 +833,7 @@ void pa_deinterleave(const void *src, void *dst[], unsigned channels, size_t ss,
         d = dst[c];
 
         for (j = 0; j < n; j ++) {
-            oil_memcpy(d, s, (int) ss);
+            memcpy(d, s, (int) ss);
             s = (uint8_t*) s + fs;
             d = (uint8_t*) d + ss;
         }
@@ -1227,10 +942,15 @@ void pa_sample_clamp(pa_sample_format_t format, void *dst, size_t dstr, const vo
     s = src; d = dst;
 
     if (format == PA_SAMPLE_FLOAT32NE) {
+        for (; n > 0; n--) {
+            float f;
 
-        float minus_one = -1.0, plus_one = 1.0;
-        oil_clip_f32(d, (int) dstr, s, (int) sstr, (int) n, &minus_one, &plus_one);
+            f = *s;
+            *d = PA_CLAMP_UNLIKELY(f, -1.0f, 1.0f);
 
+            s = (const float*) ((const uint8_t*) s + sstr);
+            d = (float*) ((uint8_t*) d + dstr);
+        }
     } else {
         pa_assert(format == PA_SAMPLE_FLOAT32RE);
 
diff --git a/src/pulsecore/sample-util.h b/src/pulsecore/sample-util.h
index 6a306c11..34df5cf3 100644
--- a/src/pulsecore/sample-util.h
+++ b/src/pulsecore/sample-util.h
@@ -86,6 +86,11 @@ void pa_memchunk_dump_to_file(pa_memchunk *c, const char *fn);
 
 void pa_memchunk_sine(pa_memchunk *c, pa_mempool *pool, unsigned rate, unsigned freq);
 
+typedef void (*pa_do_volume_func_t) (void *samples, void *volumes, unsigned channels, unsigned length);
+
+pa_do_volume_func_t pa_get_volume_func(pa_sample_format_t f);
+void pa_set_volume_func(pa_sample_format_t f, pa_do_volume_func_t func);
+
 #define PA_CHANNEL_POSITION_MASK_LEFT                                   \
     (PA_CHANNEL_POSITION_MASK(PA_CHANNEL_POSITION_FRONT_LEFT)           \
      | PA_CHANNEL_POSITION_MASK(PA_CHANNEL_POSITION_REAR_LEFT)          \
diff --git a/src/pulsecore/sconv-s16le.c b/src/pulsecore/sconv-s16le.c
index 43b8cb3e..0fefdf1c 100644
--- a/src/pulsecore/sconv-s16le.c
+++ b/src/pulsecore/sconv-s16le.c
@@ -28,8 +28,6 @@
 #include <inttypes.h>
 #include <stdio.h>
 
-#include <liboil/liboilfuncs.h>
-
 #include <pulsecore/sconv.h>
 #include <pulsecore/macro.h>
 #include <pulsecore/log.h>
@@ -86,17 +84,13 @@ void pa_sconv_s16le_to_float32ne(unsigned n, const int16_t *a, float *b) {
     pa_assert(b);
 
 #if SWAP_WORDS == 1
-
     for (; n > 0; n--) {
         int16_t s = *(a++);
         *(b++) = ((float) INT16_FROM(s))/(float) 0x7FFF;
     }
-
 #else
-{
-    static const double add = 0, factor = 1.0/0x7FFF;
-    oil_scaleconv_f32_s16(b, a, (int) n, &add, &factor);
-}
+    for (; n > 0; n--)
+        *(b++) = ((float) (*(a++)))/(float) 0x7FFF;
 #endif
 }
 
@@ -105,17 +99,13 @@ void pa_sconv_s32le_to_float32ne(unsigned n, const int32_t *a, float *b) {
     pa_assert(b);
 
 #if SWAP_WORDS == 1
-
     for (; n > 0; n--) {
         int32_t s = *(a++);
         *(b++) = (float) (((double) INT32_FROM(s))/0x7FFFFFFF);
     }
-
 #else
-{
-    static const double add = 0, factor = 1.0/0x7FFFFFFF;
-    oil_scaleconv_f32_s32(b, a, (int) n, &add, &factor);
-}
+    for (; n > 0; n--)
+        *(b++) = (float) (((double) (*(a++)))/0x7FFFFFFF);
 #endif
 }
 
@@ -124,7 +114,6 @@ void pa_sconv_s16le_from_float32ne(unsigned n, const float *a, int16_t *b) {
     pa_assert(b);
 
 #if SWAP_WORDS == 1
-
     for (; n > 0; n--) {
         int16_t s;
         float v = *(a++);
@@ -133,12 +122,13 @@ void pa_sconv_s16le_from_float32ne(unsigned n, const float *a, int16_t *b) {
         s = (int16_t) lrintf(v * 0x7FFF);
         *(b++) = INT16_TO(s);
     }
-
 #else
-{
-    static const double add = 0, factor = 0x7FFF;
-    oil_scaleconv_s16_f32(b, a, (int) n, &add, &factor);
-}
+    for (; n > 0; n--) {
+        float v = *(a++);
+
+        v = PA_CLAMP_UNLIKELY(v, -1.0f, 1.f);
+        *(b++) = (int16_t) lrintf(v * 0x7FFF);
+    }
 #endif
 }
 
@@ -147,7 +137,6 @@ void pa_sconv_s32le_from_float32ne(unsigned n, const float *a, int32_t *b) {
     pa_assert(b);
 
 #if SWAP_WORDS == 1
-
     for (; n > 0; n--) {
         int32_t s;
         float v = *(a++);
@@ -156,12 +145,13 @@ void pa_sconv_s32le_from_float32ne(unsigned n, const float *a, int32_t *b) {
         s = (int32_t) lrint((double) v * (double) 0x7FFFFFFF);
         *(b++) = INT32_TO(s);
     }
-
 #else
-{
-    static const double add = 0, factor = 0x7FFFFFFF;
-    oil_scaleconv_s32_f32(b, a, (int) n, &add, &factor);
-}
+    for (; n > 0; n--) {
+        float v = *(a++);
+
+        v = PA_CLAMP_UNLIKELY(v, -1.0f, 1.0f);
+        *(b++) = (int32_t) lrint((double) v * (double) 0x7FFFFFFF);
+    }
 #endif
 }
 
diff --git a/src/pulsecore/sconv.c b/src/pulsecore/sconv.c
index d89f4283..d06d6985 100644
--- a/src/pulsecore/sconv.c
+++ b/src/pulsecore/sconv.c
@@ -27,9 +27,6 @@
 #include <stdio.h>
 #include <stdlib.h>
 
-#include <liboil/liboilfuncs.h>
-#include <liboil/liboil.h>
-
 #include <pulsecore/g711.h>
 #include <pulsecore/macro.h>
 
@@ -41,32 +38,31 @@
 
 /* u8 */
 static void u8_to_float32ne(unsigned n, const uint8_t *a, float *b) {
-    static const double add = -1, factor = 1.0/128.0;
-
     pa_assert(a);
     pa_assert(b);
 
-    oil_scaleconv_f32_u8(b, a, (int) n, &add, &factor);
+    for (; n > 0; n--, a++, b++)
+        *b = (*a * 1.0/128.0) - 1.0;
 }
 
 static void u8_from_float32ne(unsigned n, const float *a, uint8_t *b) {
-    static const double add = 128, factor = 127.0;
-
     pa_assert(a);
     pa_assert(b);
 
-    oil_scaleconv_u8_f32(b, a, (int) n, &add, &factor);
+    for (; n > 0; n--, a++, b++) {
+        float v;
+        v = (*a * 127.0) + 128.0;
+	v = PA_CLAMP_UNLIKELY (v, 0.0, 255.0);
+	*b = rint (v);
+    }
 }
 
 static void u8_to_s16ne(unsigned n, const uint8_t *a, int16_t *b) {
-    static const int16_t add = -0x80, factor = 0x100;
-
     pa_assert(a);
     pa_assert(b);
 
-    oil_conv_s16_u8(b, 2, a, 1, (int) n);
-    oil_scalaradd_s16(b, 2, b, 2, &add, (int) n);
-    oil_scalarmult_s16(b, 2, b, 2, &factor, (int) n);
+    for (; n > 0; n--, a++, b++)
+        *b = (((int16_t)*a) - 128) << 8;
 }
 
 static void u8_from_s16ne(unsigned n, const int16_t *a, uint8_t *b) {
@@ -84,7 +80,7 @@ static void float32ne_to_float32ne(unsigned n, const float *a, float *b) {
     pa_assert(a);
     pa_assert(b);
 
-    oil_memcpy(b, a, (int) (sizeof(float) * n));
+    memcpy(b, a, (int) (sizeof(float) * n));
 }
 
 static void float32re_to_float32ne(unsigned n, const float *a, float *b) {
@@ -101,7 +97,7 @@ static void s16ne_to_s16ne(unsigned n, const int16_t *a, int16_t *b) {
     pa_assert(a);
     pa_assert(b);
 
-    oil_memcpy(b, a, (int) (sizeof(int16_t) * n));
+    memcpy(b, a, (int) (sizeof(int16_t) * n));
 }
 
 static void s16re_to_s16ne(unsigned n, const int16_t *a, int16_t *b) {
@@ -188,98 +184,130 @@ static void alaw_from_s16ne(unsigned n, const int16_t *a, uint8_t *b) {
         *b = st_13linear2alaw(*a >> 3);
 }
 
+static pa_convert_func_t to_float32ne_table[] = {
+    [PA_SAMPLE_U8]        = (pa_convert_func_t) u8_to_float32ne,
+    [PA_SAMPLE_ALAW]      = (pa_convert_func_t) alaw_to_float32ne,
+    [PA_SAMPLE_ULAW]      = (pa_convert_func_t) ulaw_to_float32ne,
+    [PA_SAMPLE_S16LE]     = (pa_convert_func_t) pa_sconv_s16le_to_float32ne,
+    [PA_SAMPLE_S16BE]     = (pa_convert_func_t) pa_sconv_s16be_to_float32ne,
+    [PA_SAMPLE_S32LE]     = (pa_convert_func_t) pa_sconv_s32le_to_float32ne,
+    [PA_SAMPLE_S32BE]     = (pa_convert_func_t) pa_sconv_s32be_to_float32ne,
+    [PA_SAMPLE_S24LE]     = (pa_convert_func_t) pa_sconv_s24le_to_float32ne,
+    [PA_SAMPLE_S24BE]     = (pa_convert_func_t) pa_sconv_s24be_to_float32ne,
+    [PA_SAMPLE_S24_32LE]  = (pa_convert_func_t) pa_sconv_s24_32le_to_float32ne,
+    [PA_SAMPLE_S24_32BE]  = (pa_convert_func_t) pa_sconv_s24_32be_to_float32ne,
+    [PA_SAMPLE_FLOAT32NE] = (pa_convert_func_t) float32ne_to_float32ne,
+    [PA_SAMPLE_FLOAT32RE] = (pa_convert_func_t) float32re_to_float32ne,
+};
+
 pa_convert_func_t pa_get_convert_to_float32ne_function(pa_sample_format_t f) {
 
-    static const pa_convert_func_t table[] = {
-        [PA_SAMPLE_U8]        = (pa_convert_func_t) u8_to_float32ne,
-        [PA_SAMPLE_ALAW]      = (pa_convert_func_t) alaw_to_float32ne,
-        [PA_SAMPLE_ULAW]      = (pa_convert_func_t) ulaw_to_float32ne,
-        [PA_SAMPLE_S16LE]     = (pa_convert_func_t) pa_sconv_s16le_to_float32ne,
-        [PA_SAMPLE_S16BE]     = (pa_convert_func_t) pa_sconv_s16be_to_float32ne,
-        [PA_SAMPLE_S32LE]     = (pa_convert_func_t) pa_sconv_s32le_to_float32ne,
-        [PA_SAMPLE_S32BE]     = (pa_convert_func_t) pa_sconv_s32be_to_float32ne,
-        [PA_SAMPLE_S24LE]     = (pa_convert_func_t) pa_sconv_s24le_to_float32ne,
-        [PA_SAMPLE_S24BE]     = (pa_convert_func_t) pa_sconv_s24be_to_float32ne,
-        [PA_SAMPLE_S24_32LE]  = (pa_convert_func_t) pa_sconv_s24_32le_to_float32ne,
-        [PA_SAMPLE_S24_32BE]  = (pa_convert_func_t) pa_sconv_s24_32be_to_float32ne,
-        [PA_SAMPLE_FLOAT32NE] = (pa_convert_func_t) float32ne_to_float32ne,
-        [PA_SAMPLE_FLOAT32RE] = (pa_convert_func_t) float32re_to_float32ne,
-    };
+    pa_assert(f >= 0);
+    pa_assert(f < PA_SAMPLE_MAX);
+
+    return to_float32ne_table[f];
+}
+
+void pa_set_convert_to_float32ne_function(pa_sample_format_t f, pa_convert_func_t func) {
 
     pa_assert(f >= 0);
     pa_assert(f < PA_SAMPLE_MAX);
 
-    return table[f];
+    to_float32ne_table[f] = func;
 }
 
+static pa_convert_func_t from_float32ne_table[] = {
+    [PA_SAMPLE_U8]        = (pa_convert_func_t) u8_from_float32ne,
+    [PA_SAMPLE_S16LE]     = (pa_convert_func_t) pa_sconv_s16le_from_float32ne,
+    [PA_SAMPLE_S16BE]     = (pa_convert_func_t) pa_sconv_s16be_from_float32ne,
+    [PA_SAMPLE_S32LE]     = (pa_convert_func_t) pa_sconv_s32le_from_float32ne,
+    [PA_SAMPLE_S32BE]     = (pa_convert_func_t) pa_sconv_s32be_from_float32ne,
+    [PA_SAMPLE_S24LE]     = (pa_convert_func_t) pa_sconv_s24le_from_float32ne,
+    [PA_SAMPLE_S24BE]     = (pa_convert_func_t) pa_sconv_s24be_from_float32ne,
+    [PA_SAMPLE_S24_32LE]  = (pa_convert_func_t) pa_sconv_s24_32le_from_float32ne,
+    [PA_SAMPLE_S24_32BE]  = (pa_convert_func_t) pa_sconv_s24_32be_from_float32ne,
+    [PA_SAMPLE_FLOAT32NE] = (pa_convert_func_t) float32ne_to_float32ne,
+    [PA_SAMPLE_FLOAT32RE] = (pa_convert_func_t) float32re_to_float32ne,
+    [PA_SAMPLE_ALAW]      = (pa_convert_func_t) alaw_from_float32ne,
+    [PA_SAMPLE_ULAW]      = (pa_convert_func_t) ulaw_from_float32ne
+};
+
 pa_convert_func_t pa_get_convert_from_float32ne_function(pa_sample_format_t f) {
 
-    static const pa_convert_func_t table[] = {
-        [PA_SAMPLE_U8]        = (pa_convert_func_t) u8_from_float32ne,
-        [PA_SAMPLE_S16LE]     = (pa_convert_func_t) pa_sconv_s16le_from_float32ne,
-        [PA_SAMPLE_S16BE]     = (pa_convert_func_t) pa_sconv_s16be_from_float32ne,
-        [PA_SAMPLE_S32LE]     = (pa_convert_func_t) pa_sconv_s32le_from_float32ne,
-        [PA_SAMPLE_S32BE]     = (pa_convert_func_t) pa_sconv_s32be_from_float32ne,
-        [PA_SAMPLE_S24LE]     = (pa_convert_func_t) pa_sconv_s24le_from_float32ne,
-        [PA_SAMPLE_S24BE]     = (pa_convert_func_t) pa_sconv_s24be_from_float32ne,
-        [PA_SAMPLE_S24_32LE]  = (pa_convert_func_t) pa_sconv_s24_32le_from_float32ne,
-        [PA_SAMPLE_S24_32BE]  = (pa_convert_func_t) pa_sconv_s24_32be_from_float32ne,
-        [PA_SAMPLE_FLOAT32NE] = (pa_convert_func_t) float32ne_to_float32ne,
-        [PA_SAMPLE_FLOAT32RE] = (pa_convert_func_t) float32re_to_float32ne,
-        [PA_SAMPLE_ALAW]      = (pa_convert_func_t) alaw_from_float32ne,
-        [PA_SAMPLE_ULAW]      = (pa_convert_func_t) ulaw_from_float32ne
-    };
+    pa_assert(f >= 0);
+    pa_assert(f < PA_SAMPLE_MAX);
+
+    return from_float32ne_table[f];
+}
+
+void pa_set_convert_from_float32ne_function(pa_sample_format_t f, pa_convert_func_t func) {
 
     pa_assert(f >= 0);
     pa_assert(f < PA_SAMPLE_MAX);
 
-    return table[f];
+    from_float32ne_table[f] = func;
 }
 
+static pa_convert_func_t to_s16ne_table[] = {
+    [PA_SAMPLE_U8]        = (pa_convert_func_t) u8_to_s16ne,
+    [PA_SAMPLE_S16NE]     = (pa_convert_func_t) s16ne_to_s16ne,
+    [PA_SAMPLE_S16RE]     = (pa_convert_func_t) s16re_to_s16ne,
+    [PA_SAMPLE_FLOAT32BE] = (pa_convert_func_t) pa_sconv_float32be_to_s16ne,
+    [PA_SAMPLE_FLOAT32LE] = (pa_convert_func_t) pa_sconv_float32le_to_s16ne,
+    [PA_SAMPLE_S32BE]     = (pa_convert_func_t) pa_sconv_s32be_to_s16ne,
+    [PA_SAMPLE_S32LE]     = (pa_convert_func_t) pa_sconv_s32le_to_s16ne,
+    [PA_SAMPLE_S24BE]     = (pa_convert_func_t) pa_sconv_s24be_to_s16ne,
+    [PA_SAMPLE_S24LE]     = (pa_convert_func_t) pa_sconv_s24le_to_s16ne,
+    [PA_SAMPLE_S24_32BE]  = (pa_convert_func_t) pa_sconv_s24_32be_to_s16ne,
+    [PA_SAMPLE_S24_32LE]  = (pa_convert_func_t) pa_sconv_s24_32le_to_s16ne,
+    [PA_SAMPLE_ALAW]      = (pa_convert_func_t) alaw_to_s16ne,
+    [PA_SAMPLE_ULAW]      = (pa_convert_func_t) ulaw_to_s16ne
+};
+
 pa_convert_func_t pa_get_convert_to_s16ne_function(pa_sample_format_t f) {
 
-    static const pa_convert_func_t table[] = {
-        [PA_SAMPLE_U8]        = (pa_convert_func_t) u8_to_s16ne,
-        [PA_SAMPLE_S16NE]     = (pa_convert_func_t) s16ne_to_s16ne,
-        [PA_SAMPLE_S16RE]     = (pa_convert_func_t) s16re_to_s16ne,
-        [PA_SAMPLE_FLOAT32BE] = (pa_convert_func_t) pa_sconv_float32be_to_s16ne,
-        [PA_SAMPLE_FLOAT32LE] = (pa_convert_func_t) pa_sconv_float32le_to_s16ne,
-        [PA_SAMPLE_S32BE]     = (pa_convert_func_t) pa_sconv_s32be_to_s16ne,
-        [PA_SAMPLE_S32LE]     = (pa_convert_func_t) pa_sconv_s32le_to_s16ne,
-        [PA_SAMPLE_S24BE]     = (pa_convert_func_t) pa_sconv_s24be_to_s16ne,
-        [PA_SAMPLE_S24LE]     = (pa_convert_func_t) pa_sconv_s24le_to_s16ne,
-        [PA_SAMPLE_S24_32BE]  = (pa_convert_func_t) pa_sconv_s24_32be_to_s16ne,
-        [PA_SAMPLE_S24_32LE]  = (pa_convert_func_t) pa_sconv_s24_32le_to_s16ne,
-        [PA_SAMPLE_ALAW]      = (pa_convert_func_t) alaw_to_s16ne,
-        [PA_SAMPLE_ULAW]      = (pa_convert_func_t) ulaw_to_s16ne
-    };
+    pa_assert(f >= 0);
+    pa_assert(f < PA_SAMPLE_MAX);
+
+    return to_s16ne_table[f];
+}
+
+void pa_set_convert_to_s16ne_function(pa_sample_format_t f, pa_convert_func_t func) {
 
     pa_assert(f >= 0);
     pa_assert(f < PA_SAMPLE_MAX);
 
-    return table[f];
+    to_s16ne_table[f] = func;
 }
 
+static pa_convert_func_t from_s16ne_table[] = {
+    [PA_SAMPLE_U8]        = (pa_convert_func_t) u8_from_s16ne,
+    [PA_SAMPLE_S16NE]     = (pa_convert_func_t) s16ne_to_s16ne,
+    [PA_SAMPLE_S16RE]     = (pa_convert_func_t) s16re_to_s16ne,
+    [PA_SAMPLE_FLOAT32BE] = (pa_convert_func_t) pa_sconv_float32be_from_s16ne,
+    [PA_SAMPLE_FLOAT32LE] = (pa_convert_func_t) pa_sconv_float32le_from_s16ne,
+    [PA_SAMPLE_S32BE]     = (pa_convert_func_t) pa_sconv_s32be_from_s16ne,
+    [PA_SAMPLE_S32LE]     = (pa_convert_func_t) pa_sconv_s32le_from_s16ne,
+    [PA_SAMPLE_S24BE]     = (pa_convert_func_t) pa_sconv_s24be_from_s16ne,
+    [PA_SAMPLE_S24LE]     = (pa_convert_func_t) pa_sconv_s24le_from_s16ne,
+    [PA_SAMPLE_S24_32BE]  = (pa_convert_func_t) pa_sconv_s24_32be_from_s16ne,
+    [PA_SAMPLE_S24_32LE]  = (pa_convert_func_t) pa_sconv_s24_32le_from_s16ne,
+    [PA_SAMPLE_ALAW]      = (pa_convert_func_t) alaw_from_s16ne,
+    [PA_SAMPLE_ULAW]      = (pa_convert_func_t) ulaw_from_s16ne,
+};
+
 pa_convert_func_t pa_get_convert_from_s16ne_function(pa_sample_format_t f) {
 
-    static const pa_convert_func_t table[] = {
-        [PA_SAMPLE_U8]        = (pa_convert_func_t) u8_from_s16ne,
-        [PA_SAMPLE_S16NE]     = (pa_convert_func_t) s16ne_to_s16ne,
-        [PA_SAMPLE_S16RE]     = (pa_convert_func_t) s16re_to_s16ne,
-        [PA_SAMPLE_FLOAT32BE] = (pa_convert_func_t) pa_sconv_float32be_from_s16ne,
-        [PA_SAMPLE_FLOAT32LE] = (pa_convert_func_t) pa_sconv_float32le_from_s16ne,
-        [PA_SAMPLE_S32BE]     = (pa_convert_func_t) pa_sconv_s32be_from_s16ne,
-        [PA_SAMPLE_S32LE]     = (pa_convert_func_t) pa_sconv_s32le_from_s16ne,
-        [PA_SAMPLE_S24BE]     = (pa_convert_func_t) pa_sconv_s24be_from_s16ne,
-        [PA_SAMPLE_S24LE]     = (pa_convert_func_t) pa_sconv_s24le_from_s16ne,
-        [PA_SAMPLE_S24_32BE]  = (pa_convert_func_t) pa_sconv_s24_32be_from_s16ne,
-        [PA_SAMPLE_S24_32LE]  = (pa_convert_func_t) pa_sconv_s24_32le_from_s16ne,
-        [PA_SAMPLE_ALAW]      = (pa_convert_func_t) alaw_from_s16ne,
-        [PA_SAMPLE_ULAW]      = (pa_convert_func_t) ulaw_from_s16ne,
-    };
+    pa_assert(f >= 0);
+    pa_assert(f < PA_SAMPLE_MAX);
+
+    return from_s16ne_table[f];
+}
+
+void pa_set_convert_from_s16ne_function(pa_sample_format_t f, pa_convert_func_t func) {
 
     pa_assert(f >= 0);
     pa_assert(f < PA_SAMPLE_MAX);
 
-    return table[f];
+    from_s16ne_table[f] = func;
 }
diff --git a/src/pulsecore/sconv.h b/src/pulsecore/sconv.h
index b00a16a4..cd937559 100644
--- a/src/pulsecore/sconv.h
+++ b/src/pulsecore/sconv.h
@@ -33,4 +33,10 @@ pa_convert_func_t pa_get_convert_from_float32ne_function(pa_sample_format_t f) P
 pa_convert_func_t pa_get_convert_to_s16ne_function(pa_sample_format_t f) PA_GCC_PURE;
 pa_convert_func_t pa_get_convert_from_s16ne_function(pa_sample_format_t f) PA_GCC_PURE;
 
+void pa_set_convert_to_float32ne_function(pa_sample_format_t f, pa_convert_func_t func);
+void pa_set_convert_from_float32ne_function(pa_sample_format_t f, pa_convert_func_t func);
+
+void pa_set_convert_to_s16ne_function(pa_sample_format_t f, pa_convert_func_t func);
+void pa_set_convert_from_s16ne_function(pa_sample_format_t f, pa_convert_func_t func);
+
 #endif
diff --git a/src/pulsecore/sink-input.c b/src/pulsecore/sink-input.c
index a29334f9..0ad95e6f 100644
--- a/src/pulsecore/sink-input.c
+++ b/src/pulsecore/sink-input.c
@@ -44,7 +44,7 @@
 #define MEMBLOCKQ_MAXLENGTH (32*1024*1024)
 #define CONVERT_BUFFER_LENGTH (PA_PAGE_SIZE)
 
-static PA_DEFINE_CHECK_TYPE(pa_sink_input, pa_msgobject);
+PA_DEFINE_PUBLIC_CLASS(pa_sink_input, pa_msgobject);
 
 static void sink_input_free(pa_object *o);
 static void set_real_ratio(pa_sink_input *i, const pa_cvolume *v);
@@ -126,6 +126,8 @@ static void reset_callbacks(pa_sink_input *i) {
     i->state_change = NULL;
     i->may_move_to = NULL;
     i->send_event = NULL;
+    i->volume_changed = NULL;
+    i->mute_changed = NULL;
 }
 
 /* Called from main context */
@@ -485,7 +487,10 @@ static void sink_input_free(pa_object *o) {
 
     pa_log_info("Freeing input %u \"%s\"", i->index, pa_strnull(pa_proplist_gets(i->proplist, PA_PROP_MEDIA_NAME)));
 
-    pa_assert(!i->thread_info.attached);
+    /* Side note: this function must be able to destruct properly any
+     * kind of sink input in any state, even those which are
+     * "half-moved" or are connected to sinks that have no asyncmsgq
+     * and are hence half-destructed themselves! */
 
     if (i->thread_info.render_memblockq)
         pa_memblockq_free(i->thread_info.render_memblockq);
@@ -968,7 +973,10 @@ void pa_sink_input_set_volume(pa_sink_input *i, const pa_cvolume *volume, pa_boo
         pa_assert_se(pa_asyncmsgq_send(i->sink->asyncmsgq, PA_MSGOBJECT(i), PA_SINK_INPUT_MESSAGE_SET_SOFT_VOLUME, NULL, 0, NULL) == 0);
     }
 
-    /* The virtual volume changed, let's tell people so */
+    /* The volume changed, let's tell people so */
+    if (i->volume_changed)
+        i->volume_changed(i);
+
     pa_subscription_post(i->core, PA_SUBSCRIPTION_EVENT_SINK_INPUT|PA_SUBSCRIPTION_EVENT_CHANGE, i->index);
 }
 
@@ -999,6 +1007,11 @@ void pa_sink_input_set_mute(pa_sink_input *i, pa_bool_t mute, pa_bool_t save) {
     i->save_muted = save;
 
     pa_assert_se(pa_asyncmsgq_send(i->sink->asyncmsgq, PA_MSGOBJECT(i), PA_SINK_INPUT_MESSAGE_SET_SOFT_MUTE, NULL, 0, NULL) == 0);
+
+    /* The mute status changed, let's tell people so */
+    if (i->mute_changed)
+        i->mute_changed(i);
+
     pa_subscription_post(i->core, PA_SUBSCRIPTION_EVENT_SINK_INPUT|PA_SUBSCRIPTION_EVENT_CHANGE, i->index);
 }
 
@@ -1263,6 +1276,10 @@ int pa_sink_input_finish_move(pa_sink_input *i, pa_sink *dest, pa_bool_t save) {
 
     /* Notify everyone */
     pa_hook_fire(&i->core->hooks[PA_CORE_HOOK_SINK_INPUT_MOVE_FINISH], i);
+
+    if (i->volume_changed)
+        i->volume_changed(i);
+
     pa_subscription_post(i->core, PA_SUBSCRIPTION_EVENT_SINK_INPUT|PA_SUBSCRIPTION_EVENT_CHANGE, i->index);
 
     return 0;
diff --git a/src/pulsecore/sink-input.h b/src/pulsecore/sink-input.h
index ea0f8c0e..fe6cf75c 100644
--- a/src/pulsecore/sink-input.h
+++ b/src/pulsecore/sink-input.h
@@ -192,8 +192,16 @@ struct pa_sink_input {
     pa_bool_t (*may_move_to) (pa_sink_input *i, pa_sink *s); /* may be NULL */
 
     /* If non-NULL this function is used to dispatch asynchronous
-     * control events. */
-    void (*send_event)(pa_sink_input *i, const char *event, pa_proplist* data);
+     * control events. Called from main context. */
+    void (*send_event)(pa_sink_input *i, const char *event, pa_proplist* data); /* may be NULL */
+
+    /* If non-NULL this function is called whenever the sink input
+     * volume changes. Called from main context */
+    void (*volume_changed)(pa_sink_input *i); /* may be NULL */
+
+    /* If non-NULL this function is called whenever the sink input
+     * mute status changes. Called from main context */
+    void (*mute_changed)(pa_sink_input *i); /* may be NULL */
 
     struct {
         pa_sink_input_state_t state;
@@ -227,7 +235,7 @@ struct pa_sink_input {
     void *userdata;
 };
 
-PA_DECLARE_CLASS(pa_sink_input);
+PA_DECLARE_PUBLIC_CLASS(pa_sink_input);
 #define PA_SINK_INPUT(o) pa_sink_input_cast(o)
 
 enum {
diff --git a/src/pulsecore/sink.c b/src/pulsecore/sink.c
index 1cce8e6b..5cec7747 100644
--- a/src/pulsecore/sink.c
+++ b/src/pulsecore/sink.c
@@ -52,7 +52,7 @@
 #define ABSOLUTE_MAX_LATENCY (10*PA_USEC_PER_SEC)
 #define DEFAULT_FIXED_LATENCY (250*PA_USEC_PER_MSEC)
 
-static PA_DEFINE_CHECK_TYPE(pa_sink, pa_msgobject);
+PA_DEFINE_PUBLIC_CLASS(pa_sink, pa_msgobject);
 
 static void sink_free(pa_object *s);
 
@@ -1380,9 +1380,14 @@ static void propagate_reference_volume(pa_sink *s) {
         pa_cvolume_remap(&remapped, &s->channel_map, &i->channel_map);
         pa_sw_cvolume_multiply(&i->volume, &remapped, &i->reference_ratio);
 
-        /* The reference volume changed, let's tell people so */
-        if (!pa_cvolume_equal(&old_volume, &i->volume))
+        /* The volume changed, let's tell people so */
+        if (!pa_cvolume_equal(&old_volume, &i->volume)) {
+
+            if (i->volume_changed)
+                i->volume_changed(i);
+
             pa_subscription_post(i->core, PA_SUBSCRIPTION_EVENT_SINK_INPUT|PA_SUBSCRIPTION_EVENT_CHANGE, i->index);
+        }
     }
 }
 
@@ -1522,8 +1527,13 @@ static void propagate_real_volume(pa_sink *s, const pa_cvolume *old_real_volume)
             pa_sw_cvolume_multiply(&i->volume, &remapped, &i->reference_ratio);
 
             /* Notify if something changed */
-            if (!pa_cvolume_equal(&old_volume, &i->volume))
+            if (!pa_cvolume_equal(&old_volume, &i->volume)) {
+
+                if (i->volume_changed)
+                    i->volume_changed(i);
+
                 pa_subscription_post(i->core, PA_SUBSCRIPTION_EVENT_SINK_INPUT|PA_SUBSCRIPTION_EVENT_CHANGE, i->index);
+            }
         }
     }
 
diff --git a/src/pulsecore/sink.h b/src/pulsecore/sink.h
index 936d1c2a..b5284b71 100644
--- a/src/pulsecore/sink.h
+++ b/src/pulsecore/sink.h
@@ -191,7 +191,7 @@ struct pa_sink {
     void *userdata;
 };
 
-PA_DECLARE_CLASS(pa_sink);
+PA_DECLARE_PUBLIC_CLASS(pa_sink);
 #define PA_SINK(s) (pa_sink_cast(s))
 
 typedef enum pa_sink_message {
diff --git a/src/pulsecore/sound-file-stream.c b/src/pulsecore/sound-file-stream.c
index 502e5c69..f41c53f3 100644
--- a/src/pulsecore/sound-file-stream.c
+++ b/src/pulsecore/sound-file-stream.c
@@ -64,9 +64,8 @@ enum {
     FILE_STREAM_MESSAGE_UNLINK
 };
 
-PA_DECLARE_CLASS(file_stream);
+PA_DEFINE_PRIVATE_CLASS(file_stream, pa_msgobject);
 #define FILE_STREAM(o) (file_stream_cast(o))
-static PA_DEFINE_CHECK_TYPE(file_stream, pa_msgobject);
 
 /* Called from main context */
 static void file_stream_unlink(file_stream *u) {
diff --git a/src/pulsecore/source-output.c b/src/pulsecore/source-output.c
index 3803a6cc..43733400 100644
--- a/src/pulsecore/source-output.c
+++ b/src/pulsecore/source-output.c
@@ -41,7 +41,7 @@
 
 #define MEMBLOCKQ_MAXLENGTH (32*1024*1024)
 
-static PA_DEFINE_CHECK_TYPE(pa_source_output, pa_msgobject);
+PA_DEFINE_PUBLIC_CLASS(pa_source_output, pa_msgobject);
 
 static void source_output_free(pa_object* mo);
 
@@ -359,8 +359,6 @@ static void source_output_free(pa_object* mo) {
 
     pa_log_info("Freeing output %u \"%s\"", o->index, pa_strnull(pa_proplist_gets(o->proplist, PA_PROP_MEDIA_NAME)));
 
-    pa_assert(!o->thread_info.attached);
-
     if (o->thread_info.delay_memblockq)
         pa_memblockq_free(o->thread_info.delay_memblockq);
 
diff --git a/src/pulsecore/source-output.h b/src/pulsecore/source-output.h
index a70a3fdb..aca9ddf2 100644
--- a/src/pulsecore/source-output.h
+++ b/src/pulsecore/source-output.h
@@ -182,7 +182,7 @@ struct pa_source_output {
     void *userdata;
 };
 
-PA_DECLARE_CLASS(pa_source_output);
+PA_DECLARE_PUBLIC_CLASS(pa_source_output);
 #define PA_SOURCE_OUTPUT(o) pa_source_output_cast(o)
 
 enum {
diff --git a/src/pulsecore/source.c b/src/pulsecore/source.c
index 8aa07f5e..3026654e 100644
--- a/src/pulsecore/source.c
+++ b/src/pulsecore/source.c
@@ -46,7 +46,7 @@
 #define ABSOLUTE_MAX_LATENCY (10*PA_USEC_PER_SEC)
 #define DEFAULT_FIXED_LATENCY (250*PA_USEC_PER_MSEC)
 
-static PA_DEFINE_CHECK_TYPE(pa_source, pa_msgobject);
+PA_DEFINE_PUBLIC_CLASS(pa_source, pa_msgobject);
 
 static void source_free(pa_object *o);
 
diff --git a/src/pulsecore/source.h b/src/pulsecore/source.h
index 7b3e4953..df3f99df 100644
--- a/src/pulsecore/source.h
+++ b/src/pulsecore/source.h
@@ -158,7 +158,7 @@ struct pa_source {
     void *userdata;
 };
 
-PA_DECLARE_CLASS(pa_source);
+PA_DECLARE_PUBLIC_CLASS(pa_source);
 #define PA_SOURCE(s) pa_source_cast(s)
 
 typedef enum pa_source_message {
diff --git a/src/pulsecore/svolume_arm.c b/src/pulsecore/svolume_arm.c
new file mode 100644
index 00000000..5bd1448f
--- /dev/null
+++ b/src/pulsecore/svolume_arm.c
@@ -0,0 +1,195 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk>
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2.1 of the License,
+  or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <pulse/timeval.h>
+#include <pulsecore/random.h>
+#include <pulsecore/macro.h>
+#include <pulsecore/g711.h>
+#include <pulsecore/core-util.h>
+
+#include "cpu-arm.h"
+
+#include "sample-util.h"
+#include "endianmacros.h"
+
+#if defined (__arm__)
+
+#define MOD_INC() \
+    " subs  r0, r6, %2              \n\t" \
+    " addcs r0, %1                  \n\t" \
+    " movcs r6, r0                  \n\t"
+
+static void
+pa_volume_s16ne_arm (int16_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    int32_t *ve;
+
+    channels = PA_MAX (4U, channels);
+    ve = volumes + channels;
+
+    __asm__ __volatile__ (
+        " mov r6, %1                      \n\t"
+        " 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"
+
+        " smulwb r0, r0, r2               \n\t"
+        " ssat r0, #16, r0                \n\t"
+
+        " strh r0, [%0], #2               \n\t"
+
+        MOD_INC()
+
+        "2:                               \n\t"
+        " mov %3, %3, LSR #1              \n\t"
+        " tst %3, #1                      \n\t" /* check for odd samples */
+        " beq  4f                         \n\t"
+
+        "3:                               \n\t"
+        " ldrd r2, [r6], #8               \n\t" /* 2 samples at a time */
+        " ldr  r0, [%0]                   \n\t"
+
+        " smulwt r2, r2, r0               \n\t"
+        " smulwb r3, r3, r0               \n\t"
+
+        " ssat r2, #16, r2                \n\t"
+        " ssat r3, #16, r3                \n\t"
+
+        " pkhbt r0, r3, r2, LSL #16       \n\t"
+        " str  r0, [%0], #4               \n\t"
+
+        MOD_INC()
+
+        "4:                               \n\t"
+        " movs %3, %3, LSR #1             \n\t"
+        " beq  6f                         \n\t"
+
+        "5:                               \n\t"
+        " ldrd r2, [r6], #8               \n\t" /* 4 samples at a time */
+        " ldrd r4, [r6], #8               \n\t"
+        " ldrd r0, [%0]                   \n\t"
+
+        " smulwt r2, r2, r0               \n\t"
+        " smulwb r3, r3, r0               \n\t"
+        " smulwt r4, r4, r1               \n\t"
+        " smulwb r5, r5, r1               \n\t"
+
+        " ssat r2, #16, r2                \n\t"
+        " ssat r3, #16, r3                \n\t"
+        " ssat r4, #16, r4                \n\t"
+        " ssat r5, #16, r5                \n\t"
+
+        " pkhbt r0, r3, r2, LSL #16       \n\t"
+        " pkhbt r1, r5, r4, LSL #16       \n\t"
+        " strd  r0, [%0], #8              \n\t"
+
+        MOD_INC()
+
+        " subs %3, %3, #1                 \n\t"
+        " bne 5b                          \n\t"
+        "6:                               \n\t"
+
+        : "+r" (samples), "+r" (volumes), "+r" (ve), "+r" (length)
+        :
+        : "r6", "r5", "r4", "r3", "r2", "r1", "r0", "cc"
+    );
+}
+
+#undef RUN_TEST
+
+#ifdef RUN_TEST
+#define CHANNELS 2
+#define SAMPLES 1023
+#define TIMES 1000
+#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;
+
+    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] = rand() >> 1;
+    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]);
+        }
+    }
+
+    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();
+    pa_log_info("ARM: %llu usec.", (long long unsigned int) (stop - start));
+
+    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();
+    pa_log_info("ref: %llu usec.", (long long unsigned int) (stop - start));
+}
+#endif
+
+#endif /* defined (__arm__) */
+
+
+void pa_volume_func_init_arm (pa_cpu_arm_flag_t flags) {
+#if defined (__arm__)
+    pa_log_info("Initialising ARM optimized 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__) */
+}
diff --git a/src/pulsecore/svolume_c.c b/src/pulsecore/svolume_c.c
new file mode 100644
index 00000000..5fc052b8
--- /dev/null
+++ b/src/pulsecore/svolume_c.c
@@ -0,0 +1,335 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2006 Pierre Ossman <ossman@cendio.se> for Cendio AB
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2.1 of the License,
+  or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+
+#include <pulsecore/macro.h>
+#include <pulsecore/g711.h>
+#include <pulsecore/core-util.h>
+
+#include "sample-util.h"
+#include "endianmacros.h"
+
+static void
+pa_volume_u8_c (uint8_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+
+    for (channel = 0; length; length--) {
+        int32_t t, hi, lo;
+
+        hi = volumes[channel] >> 16;
+        lo = volumes[channel] & 0xFFFF;
+
+        t = (int32_t) *samples - 0x80;
+        t = ((t * lo) >> 16) + (t * hi);
+        t = PA_CLAMP_UNLIKELY(t, -0x80, 0x7F);
+        *samples++ = (uint8_t) (t + 0x80);
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_alaw_c (uint8_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+
+    for (channel = 0; length; length--) {
+        int32_t t, hi, lo;
+
+        hi = volumes[channel] >> 16;
+        lo = volumes[channel] & 0xFFFF;
+
+        t = (int32_t) st_alaw2linear16(*samples);
+        t = ((t * lo) >> 16) + (t * hi);
+        t = PA_CLAMP_UNLIKELY(t, -0x8000, 0x7FFF);
+        *samples++ = (uint8_t) st_13linear2alaw((int16_t) t >> 3);
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_ulaw_c (uint8_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+
+    for (channel = 0; length; length--) {
+        int32_t t, hi, lo;
+
+        hi = volumes[channel] >> 16;
+        lo = volumes[channel] & 0xFFFF;
+
+        t = (int32_t) st_ulaw2linear16(*samples);
+        t = ((t * lo) >> 16) + (t * hi);
+        t = PA_CLAMP_UNLIKELY(t, -0x8000, 0x7FFF);
+        *samples++ = (uint8_t) st_14linear2ulaw((int16_t) t >> 2);
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_s16ne_c (int16_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+
+    length /= sizeof (int16_t);
+
+    for (channel = 0; length; length--) {
+        int32_t t, hi, lo;
+
+        /* Multiplying the 32bit volume factor with the 16bit
+         * sample might result in an 48bit value. We want to
+         * do without 64 bit integers and hence do the
+         * multiplication independantly for the HI and LO part
+         * of the volume. */
+
+        hi = volumes[channel] >> 16;
+        lo = volumes[channel] & 0xFFFF;
+
+        t = (int32_t)(*samples);
+        t = ((t * lo) >> 16) + (t * hi);
+        t = PA_CLAMP_UNLIKELY(t, -0x8000, 0x7FFF);
+        *samples++ = (int16_t) t;
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_s16re_c (int16_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+
+    length /= sizeof (int16_t);
+
+    for (channel = 0; length; length--) {
+        int32_t t, hi, lo;
+
+        hi = volumes[channel] >> 16;
+        lo = volumes[channel] & 0xFFFF;
+
+        t = (int32_t) PA_INT16_SWAP(*samples);
+        t = ((t * lo) >> 16) + (t * hi);
+        t = PA_CLAMP_UNLIKELY(t, -0x8000, 0x7FFF);
+        *samples++ = PA_INT16_SWAP((int16_t) t);
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_float32ne_c (float *samples, float *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+
+    length /= sizeof (float);
+
+    for (channel = 0; length; length--) {
+        *samples++ *= volumes[channel];
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_float32re_c (float *samples, float *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+
+    length /= sizeof (float);
+
+    for (channel = 0; length; length--) {
+        float t;
+
+        t = PA_FLOAT32_SWAP(*samples);
+        t *= volumes[channel];
+        *samples++ = PA_FLOAT32_SWAP(t);
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_s32ne_c (int32_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+
+    length /= sizeof (int32_t);
+
+    for (channel = 0; length; length--) {
+        int64_t t;
+
+        t = (int64_t)(*samples);
+        t = (t * volumes[channel]) >> 16;
+        t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
+        *samples++ = (int32_t) t;
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_s32re_c (int32_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+
+    length /= sizeof (int32_t);
+
+    for (channel = 0; length; length--) {
+        int64_t t;
+
+        t = (int64_t) PA_INT32_SWAP(*samples);
+        t = (t * volumes[channel]) >> 16;
+        t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
+        *samples++ = PA_INT32_SWAP((int32_t) t);
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_s24ne_c (uint8_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+    uint8_t *e;
+
+    e = samples + length;
+
+    for (channel = 0; samples < e; samples += 3) {
+        int64_t t;
+
+        t = (int64_t)((int32_t) (PA_READ24NE(samples) << 8));
+        t = (t * volumes[channel]) >> 16;
+        t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
+        PA_WRITE24NE(samples, ((uint32_t) (int32_t) t) >> 8);
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_s24re_c (uint8_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+    uint8_t *e;
+
+    e = samples + length;
+
+    for (channel = 0; samples < e; samples += 3) {
+        int64_t t;
+
+        t = (int64_t)((int32_t) (PA_READ24RE(samples) << 8));
+        t = (t * volumes[channel]) >> 16;
+        t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
+        PA_WRITE24RE(samples, ((uint32_t) (int32_t) t) >> 8);
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_s24_32ne_c (uint32_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+
+    length /= sizeof (uint32_t);
+
+    for (channel = 0; length; length--) {
+        int64_t t;
+
+        t = (int64_t) ((int32_t) (*samples << 8));
+        t = (t * volumes[channel]) >> 16;
+        t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
+        *samples++ = ((uint32_t) ((int32_t) t)) >> 8;
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static void
+pa_volume_s24_32re_c (uint32_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    unsigned channel;
+
+    length /= sizeof (uint32_t);
+
+    for (channel = 0; length; length--) {
+        int64_t t;
+
+        t = (int64_t) ((int32_t) (PA_UINT32_SWAP(*samples) << 8));
+        t = (t * volumes[channel]) >> 16;
+        t = PA_CLAMP_UNLIKELY(t, -0x80000000LL, 0x7FFFFFFFLL);
+        *samples++ = PA_UINT32_SWAP(((uint32_t) ((int32_t) t)) >> 8);
+
+        if (PA_UNLIKELY(++channel >= channels))
+            channel = 0;
+    }
+}
+
+static pa_do_volume_func_t do_volume_table[] =
+{
+    [PA_SAMPLE_U8]        = (pa_do_volume_func_t) pa_volume_u8_c,
+    [PA_SAMPLE_ALAW]      = (pa_do_volume_func_t) pa_volume_alaw_c,
+    [PA_SAMPLE_ULAW]      = (pa_do_volume_func_t) pa_volume_ulaw_c,
+    [PA_SAMPLE_S16NE]     = (pa_do_volume_func_t) pa_volume_s16ne_c,
+    [PA_SAMPLE_S16RE]     = (pa_do_volume_func_t) pa_volume_s16re_c,
+    [PA_SAMPLE_FLOAT32NE] = (pa_do_volume_func_t) pa_volume_float32ne_c,
+    [PA_SAMPLE_FLOAT32RE] = (pa_do_volume_func_t) pa_volume_float32re_c,
+    [PA_SAMPLE_S32NE]     = (pa_do_volume_func_t) pa_volume_s32ne_c,
+    [PA_SAMPLE_S32RE]     = (pa_do_volume_func_t) pa_volume_s32re_c,
+    [PA_SAMPLE_S24NE]     = (pa_do_volume_func_t) pa_volume_s24ne_c,
+    [PA_SAMPLE_S24RE]     = (pa_do_volume_func_t) pa_volume_s24re_c,
+    [PA_SAMPLE_S24_32NE]  = (pa_do_volume_func_t) pa_volume_s24_32ne_c,
+    [PA_SAMPLE_S24_32RE]  = (pa_do_volume_func_t) pa_volume_s24_32re_c
+};
+
+pa_do_volume_func_t pa_get_volume_func(pa_sample_format_t f) {
+    pa_assert(f >= 0);
+    pa_assert(f < PA_SAMPLE_MAX);
+
+    return do_volume_table[f];
+}
+
+void pa_set_volume_func(pa_sample_format_t f, pa_do_volume_func_t func) {
+    pa_assert(f >= 0);
+    pa_assert(f < PA_SAMPLE_MAX);
+
+    do_volume_table[f] = func;
+}
diff --git a/src/pulsecore/svolume_mmx.c b/src/pulsecore/svolume_mmx.c
new file mode 100644
index 00000000..8510b0c4
--- /dev/null
+++ b/src/pulsecore/svolume_mmx.c
@@ -0,0 +1,313 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk>
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2.1 of the License,
+  or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <pulse/timeval.h>
+#include <pulsecore/random.h>
+#include <pulsecore/macro.h>
+#include <pulsecore/g711.h>
+#include <pulsecore/core-util.h>
+
+#include "cpu-x86.h"
+
+#include "sample-util.h"
+#include "endianmacros.h"
+
+#if 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:
+ *
+ *  vl = (v & 0xffff)
+ *  vh = (v >> 16)
+ *  s = ((s * vl) >> 16) + (s * vh);
+ *
+ * For the first multiply we have to do a sign correction as we need to
+ * multiply a signed int with an unsigned int. Hacker's delight 8-3 gives a
+ * simple formula to correct the sign of the high word after the signed
+ * multiply.
+ */
+#define VOLUME_32x16(s,v)                  /* .. |   vh  |   vl  | */                   \
+      " pxor  %%mm4, %%mm4           \n\t" /* .. |    0  |    0  | */                   \
+      " punpcklwd %%mm4, "#s"        \n\t" /* .. |    0  |   p0  | */                   \
+      " pcmpgtw "#v", %%mm4          \n\t" /* .. |    0  | s(vl) | */                   \
+      " pand "#s", %%mm4             \n\t" /* .. |    0  |  (p0) |  (vl >> 15) & p */   \
+      " movq %%mm6, %%mm5            \n\t" /* .. |  ffff |   0   | */                   \
+      " pand "#v", %%mm5             \n\t" /* .. |   vh  |   0   | */                   \
+      " por %%mm5, %%mm4             \n\t" /* .. |   vh  |  (p0) | */                   \
+      " pmulhw "#s", "#v"            \n\t" /* .. |    0  | vl*p0 | */                   \
+      " paddw %%mm4, "#v"            \n\t" /* .. |   vh  | vl*p0 | vh + sign correct */ \
+      " pslld $16, "#s"              \n\t" /* .. |   p0  |    0  | */                   \
+      " por %%mm7, "#s"              \n\t" /* .. |   p0  |    1  | */                   \
+      " pmaddwd "#s", "#v"           \n\t" /* .. |    p0 * v0    | */                   \
+      " packssdw "#v", "#v"          \n\t" /* .. | p1*v1 | p0*v0 | */
+
+/* approximately advances %3 = (%3 + a) % b. This function requires that
+ * a <= b. */
+#define MOD_ADD(a,b) \
+      " add "#a", %3                 \n\t" \
+      " mov %3, %4                   \n\t" \
+      " sub "#b", %4                 \n\t" \
+      " cmovae %4, %3                \n\t"
+
+/* swap 16 bits */
+#define SWAP_16(s) \
+      " movq "#s", %%mm4             \n\t" /* .. |  h  l |  */ \
+      " psrlw $8, %%mm4              \n\t" /* .. |  0  h |  */ \
+      " psllw $8, "#s"               \n\t" /* .. |  l  0 |  */ \
+      " por %%mm4, "#s"              \n\t" /* .. |  l  h |  */
+
+/* swap 2 registers 16 bits for better pairing */
+#define SWAP_16_2(s1,s2) \
+      " movq "#s1", %%mm4            \n\t" /* .. |  h  l |  */ \
+      " movq "#s2", %%mm5            \n\t"                     \
+      " psrlw $8, %%mm4              \n\t" /* .. |  0  h |  */ \
+      " psrlw $8, %%mm5              \n\t"                     \
+      " psllw $8, "#s1"              \n\t" /* .. |  l  0 |  */ \
+      " psllw $8, "#s2"              \n\t"                     \
+      " 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)
+{
+    pa_reg_x86 channel, temp;
+
+    /* the max number of samples we process at a time, this is also the max amount
+     * we overread the volume array, which should have enough padding. */
+    channels = PA_MAX (4U, channels);
+
+    __asm__ __volatile__ (
+        " xor %3, %3                    \n\t"
+        " sar $1, %2                    \n\t" /* length /= sizeof (int16_t) */
+        " pcmpeqw %%mm6, %%mm6          \n\t" /* .. |  ffff |  ffff | */
+        " pcmpeqw %%mm7, %%mm7          \n\t" /* .. |  ffff |  ffff | */
+        " pslld  $16, %%mm6             \n\t" /* .. |  ffff |     0 | */
+        " psrld  $31, %%mm7             \n\t" /* .. |     0 |     1 | */
+
+        " test $1, %2                   \n\t" /* check for odd samples */
+        " je 2f                         \n\t"
+
+        " movd (%1, %3, 4), %%mm0       \n\t" /* |  v0h  |  v0l  | */
+        " movw (%0), %w4                \n\t" /*     ..  |  p0   | */
+        " movd %4, %%mm1                \n\t"
+        VOLUME_32x16 (%%mm1, %%mm0)
+        " movd %%mm0, %4                \n\t" /*     ..  | p0*v0 | */
+        " movw %w4, (%0)                \n\t"
+        " add $2, %0                    \n\t"
+        MOD_ADD ($1, %5)
+
+        "2:                             \n\t"
+        " sar $1, %2                    \n\t" /* prepare for processing 2 samples at a time */
+        " test $1, %2                   \n\t" /* check for odd samples */
+        " je 4f                         \n\t"
+
+        "3:                             \n\t" /* do samples in groups of 2 */
+        " movq (%1, %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 | */
+        " add $4, %0                    \n\t"
+        MOD_ADD ($2, %5)
+
+        "4:                             \n\t"
+        " sar $1, %2                    \n\t" /* prepare for processing 4 samples at a time */
+        " cmp $0, %2                    \n\t"
+        " 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  | */
+        " movd (%0), %%mm1              \n\t" /*              .. |   p1  |  p0   | */
+        " movd 4(%0), %%mm3             \n\t" /*              .. |   p3  |  p2   | */
+        VOLUME_32x16 (%%mm1, %%mm0)
+        VOLUME_32x16 (%%mm3, %%mm2)
+        " movd %%mm0, (%0)              \n\t" /*              .. | p1*v1 | p0*v0 | */
+        " movd %%mm2, 4(%0)             \n\t" /*              .. | p3*v3 | p2*v2 | */
+        " add $8, %0                    \n\t"
+        MOD_ADD ($4, %5)
+        " dec %2                        \n\t"
+        " jne 5b                        \n\t"
+
+        "6:                             \n\t"
+        " emms                          \n\t"
+
+        : "+r" (samples), "+r" (volumes), "+r" (length), "=D" ((pa_reg_x86)channel), "=&r" (temp)
+        : "r" ((pa_reg_x86)channels)
+        : "cc"
+    );
+}
+
+static void
+pa_volume_s16re_mmx (int16_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    pa_reg_x86 channel, temp;
+
+    /* the max number of samples we process at a time, this is also the max amount
+     * we overread the volume array, which should have enough padding. */
+    channels = PA_MAX (4U, channels);
+
+    __asm__ __volatile__ (
+        " xor %3, %3                    \n\t"
+        " sar $1, %2                    \n\t" /* length /= sizeof (int16_t) */
+        " pcmpeqw %%mm6, %%mm6          \n\t" /* .. |  ffff |  ffff | */
+        " pcmpeqw %%mm7, %%mm7          \n\t" /* .. |  ffff |  ffff | */
+        " pslld  $16, %%mm6             \n\t" /* .. |  ffff |     0 | */
+        " psrld  $31, %%mm7             \n\t" /* .. |     0 |     1 | */
+
+        " test $1, %2                   \n\t" /* check for odd samples */
+        " je 2f                         \n\t"
+
+        " movd (%1, %3, 4), %%mm0       \n\t" /* |  v0h  |  v0l  | */
+        " movw (%0), %w4                \n\t" /*     ..  |  p0   | */
+        " rorw $8, %w4                  \n\t"
+        " movd %4, %%mm1                \n\t"
+        VOLUME_32x16 (%%mm1, %%mm0)
+        " movd %%mm0, %4                \n\t" /*     ..  | p0*v0 | */
+        " rorw $8, %w4                  \n\t"
+        " movw %w4, (%0)                \n\t"
+        " add $2, %0                    \n\t"
+        MOD_ADD ($1, %5)
+
+        "2:                             \n\t"
+        " sar $1, %2                    \n\t" /* prepare for processing 2 samples at a time */
+        " test $1, %2                   \n\t" /* check for odd samples */
+        " je 4f                         \n\t"
+
+        "3:                             \n\t" /* do samples in groups of 2 */
+        " movq (%1, %3, 4), %%mm0       \n\t" /* |  v1h  |  v1l  |  v0h  |  v0l  | */
+        " movd (%0), %%mm1              \n\t" /*              .. |   p1  |  p0   | */
+        SWAP_16 (%%mm1)
+        VOLUME_32x16 (%%mm1, %%mm0)
+        SWAP_16 (%%mm0)
+        " movd %%mm0, (%0)              \n\t" /*              .. | p1*v1 | p0*v0 | */
+        " add $4, %0                    \n\t"
+        MOD_ADD ($2, %5)
+
+        "4:                             \n\t"
+        " sar $1, %2                    \n\t" /* prepare for processing 4 samples at a time */
+        " cmp $0, %2                    \n\t"
+        " 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  | */
+        " movd (%0), %%mm1              \n\t" /*              .. |   p1  |  p0   | */
+        " movd 4(%0), %%mm3             \n\t" /*              .. |   p3  |  p2   | */
+        SWAP_16_2 (%%mm1, %%mm3)
+        VOLUME_32x16 (%%mm1, %%mm0)
+        VOLUME_32x16 (%%mm3, %%mm2)
+        SWAP_16_2 (%%mm0, %%mm2)
+        " movd %%mm0, (%0)              \n\t" /*              .. | p1*v1 | p0*v0 | */
+        " movd %%mm2, 4(%0)             \n\t" /*              .. | p3*v3 | p2*v2 | */
+        " add $8, %0                    \n\t"
+        MOD_ADD ($4, %5)
+        " dec %2                        \n\t"
+        " jne 5b                        \n\t"
+
+        "6:                             \n\t"
+        " emms                          \n\t"
+
+        : "+r" (samples), "+r" (volumes), "+r" (length), "=D" ((pa_reg_x86)channel), "=&r" (temp)
+        : "r" ((pa_reg_x86)channels)
+        : "cc"
+    );
+}
+
+#undef RUN_TEST
+
+#ifdef RUN_TEST
+#define CHANNELS 2
+#define SAMPLES 1021
+#define TIMES 1000
+#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;
+
+    func = pa_get_volume_func (PA_SAMPLE_S16NE);
+
+    printf ("checking MMX %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] = rand() >> 1;
+    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 * %04x)\n", i, samples[i], samples_ref[i],
+                  samples_orig[i], volumes[i % CHANNELS]);
+        }
+    }
+
+    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();
+    pa_log_info("MMX: %llu usec.", (long long unsigned int)(stop - start));
+
+    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();
+    pa_log_info("ref: %llu usec.", (long long unsigned int)(stop - start));
+}
+#endif
+
+#endif /* defined (__i386__) || defined (__amd64__) */
+
+
+void pa_volume_func_init_mmx (pa_cpu_x86_flag_t flags) {
+#if defined (__i386__) || defined (__amd64__)
+    pa_log_info("Initialising MMX optimized functions.");
+
+#ifdef RUN_TEST
+    run_test ();
+#endif
+
+    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__) */
+}
diff --git a/src/pulsecore/svolume_sse.c b/src/pulsecore/svolume_sse.c
new file mode 100644
index 00000000..54af4a57
--- /dev/null
+++ b/src/pulsecore/svolume_sse.c
@@ -0,0 +1,314 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2009 Wim Taymans <wim.taymans@collabora.co.uk>
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2.1 of the License,
+  or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <pulse/timeval.h>
+#include <pulsecore/random.h>
+#include <pulsecore/macro.h>
+#include <pulsecore/g711.h>
+#include <pulsecore/core-util.h>
+
+#include "cpu-x86.h"
+
+#include "sample-util.h"
+#include "endianmacros.h"
+
+#if defined (__i386__) || defined (__amd64__)
+
+#define VOLUME_32x16(s,v)                  /* .. |   vh  |   vl  | */                   \
+      " pxor %%xmm4, %%xmm4          \n\t" /* .. |    0  |    0  | */                   \
+      " punpcklwd %%xmm4, "#s"       \n\t" /* .. |    0  |   p0  | */                   \
+      " pcmpgtw "#s", %%xmm4         \n\t" /* .. |    0  | s(p0) | */                   \
+      " pand "#v", %%xmm4            \n\t" /* .. |    0  |  (vl) | */                   \
+      " movdqa "#s", %%xmm5          \n\t"                                              \
+      " pmulhuw "#v", "#s"           \n\t" /* .. |    0  | vl*p0 | */                   \
+      " psubd %%xmm4, "#s"           \n\t" /* .. |    0  | vl*p0 | + sign correct */    \
+      " psrld $16, "#v"              \n\t" /* .. |   p0  |    0  | */                   \
+      " pmaddwd %%xmm5, "#v"         \n\t" /* .. |    p0 * vh    | */                   \
+      " paddd "#s", "#v"             \n\t" /* .. |    p0 * v0    | */                   \
+      " packssdw "#v", "#v"          \n\t" /* .. | p1*v1 | p0*v0 | */
+
+#define MOD_ADD(a,b) \
+      " add "#a", %3                 \n\t" /* channel += inc           */ \
+      " mov %3, %4                   \n\t"                                \
+      " sub "#b", %4                 \n\t" /* tmp = channel - channels */ \
+      " cmovae %4, %3                \n\t" /* if (tmp >= 0) channel = tmp  */
+
+/* swap 16 bits */
+#define SWAP_16(s) \
+      " movdqa "#s", %%xmm4          \n\t" /* .. |  h  l |  */ \
+      " psrlw $8, %%xmm4             \n\t" /* .. |  0  h |  */ \
+      " psllw $8, "#s"               \n\t" /* .. |  l  0 |  */ \
+      " por %%xmm4, "#s"             \n\t" /* .. |  l  h |  */
+
+/* swap 2 registers 16 bits for better pairing */
+#define SWAP_16_2(s1,s2) \
+      " movdqa "#s1", %%xmm4         \n\t" /* .. |  h  l |  */ \
+      " movdqa "#s2", %%xmm5         \n\t"                     \
+      " psrlw $8, %%xmm4             \n\t" /* .. |  0  h |  */ \
+      " psrlw $8, %%xmm5             \n\t"                     \
+      " psllw $8, "#s1"              \n\t" /* .. |  l  0 |  */ \
+      " psllw $8, "#s2"              \n\t"                     \
+      " por %%xmm4, "#s1"            \n\t" /* .. |  l  h |  */ \
+      " por %%xmm5, "#s2"            \n\t"
+
+static void
+pa_volume_s16ne_sse (int16_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    pa_reg_x86 channel, temp;
+
+    /* the max number of samples we process at a time, this is also the max amount
+     * we overread the volume array, which should have enough padding. */
+    channels = PA_MAX (8U, channels);
+
+    __asm__ __volatile__ (
+        " xor %3, %3                    \n\t"
+        " sar $1, %2                    \n\t" /* length /= sizeof (int16_t) */
+
+        " test $1, %2                   \n\t" /* check for odd samples */
+        " je 2f                         \n\t"
+
+        " movd (%1, %3, 4), %%xmm0      \n\t" /* |  v0h  |  v0l  | */
+        " movw (%0), %w4                \n\t" /*     ..  |   p0  | */
+        " movd %4, %%xmm1               \n\t"
+        VOLUME_32x16 (%%xmm1, %%xmm0)
+        " movd %%xmm0, %4               \n\t" /*     ..  | p0*v0 | */
+        " movw %w4, (%0)                \n\t"
+        " add $2, %0                    \n\t"
+        MOD_ADD ($1, %5)
+
+        "2:                             \n\t"
+        " sar $1, %2                    \n\t" /* prepare for processing 2 samples at a time */
+        " test $1, %2                   \n\t"
+        " je 4f                         \n\t"
+
+        "3:                             \n\t" /* do samples in groups of 2 */
+        " movq (%1, %3, 4), %%xmm0      \n\t" /* |  v1h  |  v1l  |  v0h  |  v0l  | */
+        " movd (%0), %%xmm1             \n\t" /*              .. |   p1  |  p0   | */
+        VOLUME_32x16 (%%xmm1, %%xmm0)
+        " movd %%xmm0, (%0)             \n\t" /*              .. | p1*v1 | p0*v0 | */
+        " add $4, %0                    \n\t"
+        MOD_ADD ($2, %5)
+
+        "4:                             \n\t"
+        " sar $1, %2                    \n\t" /* prepare for processing 4 samples at a time */
+        " test $1, %2                   \n\t"
+        " je 6f                         \n\t"
+
+        /* FIXME, we can do aligned access of the volume values if we can guarantee
+         * that the array is 16 bytes aligned, we probably have to do the odd values
+         * after this then. */
+        "5:                             \n\t" /* do samples in groups of 4 */
+        " movdqu (%1, %3, 4), %%xmm0    \n\t" /* |  v3h  |  v3l  ..  v0h  |  v0l  | */
+        " movq (%0), %%xmm1             \n\t" /*              .. |   p3  ..  p0   | */
+        VOLUME_32x16 (%%xmm1, %%xmm0)
+        " movq %%xmm0, (%0)             \n\t" /*              .. | p3*v3 .. p0*v0 | */
+        " add $8, %0                    \n\t"
+        MOD_ADD ($4, %5)
+
+        "6:                             \n\t"
+        " sar $1, %2                    \n\t" /* prepare for processing 8 samples at a time */
+        " cmp $0, %2                    \n\t"
+        " je 8f                         \n\t"
+
+        "7:                             \n\t" /* do samples in groups of 8 */
+        " movdqu (%1, %3, 4), %%xmm0    \n\t" /* |  v3h  |  v3l  ..  v0h  |  v0l  | */
+        " movdqu 16(%1, %3, 4), %%xmm2  \n\t" /* |  v7h  |  v7l  ..  v4h  |  v4l  | */
+        " movq (%0), %%xmm1             \n\t" /*              .. |   p3  ..  p0   | */
+        " movq 8(%0), %%xmm3            \n\t" /*              .. |   p7  ..  p4   | */
+        VOLUME_32x16 (%%xmm1, %%xmm0)
+        VOLUME_32x16 (%%xmm3, %%xmm2)
+        " movq %%xmm0, (%0)             \n\t" /*              .. | p3*v3 .. p0*v0 | */
+        " movq %%xmm2, 8(%0)            \n\t" /*              .. | p7*v7 .. p4*v4 | */
+        " add $16, %0                   \n\t"
+        MOD_ADD ($8, %5)
+        " dec %2                        \n\t"
+        " jne 7b                        \n\t"
+        "8:                             \n\t"
+
+        : "+r" (samples), "+r" (volumes), "+r" (length), "=D" (channel), "=&r" (temp)
+        : "r" ((pa_reg_x86)channels)
+        : "cc"
+    );
+}
+
+static void
+pa_volume_s16re_sse (int16_t *samples, int32_t *volumes, unsigned channels, unsigned length)
+{
+    pa_reg_x86 channel, temp;
+
+    /* the max number of samples we process at a time, this is also the max amount
+     * we overread the volume array, which should have enough padding. */
+    channels = PA_MAX (8U, channels);
+
+    __asm__ __volatile__ (
+        " xor %3, %3                    \n\t"
+        " sar $1, %2                    \n\t" /* length /= sizeof (int16_t) */
+
+        " test $1, %2                   \n\t" /* check for odd samples */
+        " je 2f                         \n\t"
+
+        " movd (%1, %3, 4), %%xmm0      \n\t" /* |  v0h  |  v0l  | */
+        " movw (%0), %w4                \n\t" /*     ..  |   p0  | */
+        " rorw $8, %w4                  \n\t"
+        " movd %4, %%xmm1               \n\t"
+        VOLUME_32x16 (%%xmm1, %%xmm0)
+        " movd %%xmm0, %4               \n\t" /*     ..  | p0*v0 | */
+        " rorw $8, %w4                  \n\t"
+        " movw %w4, (%0)                \n\t"
+        " add $2, %0                    \n\t"
+        MOD_ADD ($1, %5)
+
+        "2:                             \n\t"
+        " sar $1, %2                    \n\t" /* prepare for processing 2 samples at a time */
+        " test $1, %2                   \n\t"
+        " je 4f                         \n\t"
+
+        "3:                             \n\t" /* do samples in groups of 2 */
+        " movq (%1, %3, 4), %%xmm0      \n\t" /* |  v1h  |  v1l  |  v0h  |  v0l  | */
+        " movd (%0), %%xmm1             \n\t" /*              .. |   p1  |  p0   | */
+        SWAP_16 (%%xmm1)
+        VOLUME_32x16 (%%xmm1, %%xmm0)
+        SWAP_16 (%%xmm0)
+        " movd %%xmm0, (%0)             \n\t" /*              .. | p1*v1 | p0*v0 | */
+        " add $4, %0                    \n\t"
+        MOD_ADD ($2, %5)
+
+        "4:                             \n\t"
+        " sar $1, %2                    \n\t" /* prepare for processing 4 samples at a time */
+        " test $1, %2                   \n\t"
+        " je 6f                         \n\t"
+
+        /* FIXME, we can do aligned access of the volume values if we can guarantee
+         * that the array is 16 bytes aligned, we probably have to do the odd values
+         * after this then. */
+        "5:                             \n\t" /* do samples in groups of 4 */
+        " movdqu (%1, %3, 4), %%xmm0    \n\t" /* |  v3h  |  v3l  ..  v0h  |  v0l  | */
+        " movq (%0), %%xmm1             \n\t" /*              .. |   p3  ..  p0   | */
+        SWAP_16 (%%xmm1)
+        VOLUME_32x16 (%%xmm1, %%xmm0)
+        SWAP_16 (%%xmm0)
+        " movq %%xmm0, (%0)             \n\t" /*              .. | p3*v3 .. p0*v0 | */
+        " add $8, %0                    \n\t"
+        MOD_ADD ($4, %5)
+
+        "6:                             \n\t"
+        " sar $1, %2                    \n\t" /* prepare for processing 8 samples at a time */
+        " cmp $0, %2                    \n\t"
+        " je 8f                         \n\t"
+
+        "7:                             \n\t" /* do samples in groups of 8 */
+        " movdqu (%1, %3, 4), %%xmm0    \n\t" /* |  v3h  |  v3l  ..  v0h  |  v0l  | */
+        " movdqu 16(%1, %3, 4), %%xmm2  \n\t" /* |  v7h  |  v7l  ..  v4h  |  v4l  | */
+        " movq (%0), %%xmm1             \n\t" /*              .. |   p3  ..  p0   | */
+        " movq 8(%0), %%xmm3            \n\t" /*              .. |   p7  ..  p4   | */
+        SWAP_16_2 (%%xmm1, %%xmm3)
+        VOLUME_32x16 (%%xmm1, %%xmm0)
+        VOLUME_32x16 (%%xmm3, %%xmm2)
+        SWAP_16_2 (%%xmm0, %%xmm2)
+        " movq %%xmm0, (%0)             \n\t" /*              .. | p3*v3 .. p0*v0 | */
+        " movq %%xmm2, 8(%0)            \n\t" /*              .. | p7*v7 .. p4*v4 | */
+        " add $16, %0                   \n\t"
+        MOD_ADD ($8, %5)
+        " dec %2                        \n\t"
+        " jne 7b                        \n\t"
+        "8:                             \n\t"
+
+        : "+r" (samples), "+r" (volumes), "+r" (length), "=D" (channel), "=&r" (temp)
+        : "r" ((pa_reg_x86)channels)
+        : "cc"
+    );
+}
+
+#undef RUN_TEST
+
+#ifdef RUN_TEST
+#define CHANNELS 2
+#define SAMPLES 1021
+#define TIMES 1000
+#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;
+
+    func = pa_get_volume_func (PA_SAMPLE_S16NE);
+
+    printf ("checking SSE %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] = rand() >> 1;
+    for (padding = 0; padding < PADDING; padding++, i++)
+        volumes[i] = volumes[padding];
+
+    func (samples_ref, volumes, CHANNELS, sizeof (samples));
+    pa_volume_s16ne_sse (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]);
+        }
+    }
+
+    start = pa_rtclock_now();
+    for (j = 0; j < TIMES; j++) {
+        memcpy (samples, samples_orig, sizeof (samples));
+        pa_volume_s16ne_sse (samples, volumes, CHANNELS, sizeof (samples));
+    }
+    stop = pa_rtclock_now();
+    pa_log_info("SSE: %llu usec.", (long long unsigned int)(stop - start));
+
+    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();
+    pa_log_info("ref: %llu usec.", (long long unsigned int)(stop - start));
+}
+#endif
+#endif /* defined (__i386__) || defined (__amd64__) */
+
+void pa_volume_func_init_sse (pa_cpu_x86_flag_t flags) {
+#if defined (__i386__) || defined (__amd64__)
+    pa_log_info("Initialising SSE optimized functions.");
+
+#ifdef RUN_TEST
+    run_test ();
+#endif
+
+    pa_set_volume_func (PA_SAMPLE_S16NE,     (pa_do_volume_func_t) pa_volume_s16ne_sse);
+    pa_set_volume_func (PA_SAMPLE_S16RE,     (pa_do_volume_func_t) pa_volume_s16re_sse);
+#endif /* defined (__i386__) || defined (__amd64__) */
+}
diff --git a/src/pulsecore/usergroup.c b/src/pulsecore/usergroup.c
new file mode 100644
index 00000000..71b13bca
--- /dev/null
+++ b/src/pulsecore/usergroup.c
@@ -0,0 +1,372 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2009 Ted Percival
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as
+  published by the Free Software Foundation; either version 2.1 of the
+  License, or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  Lesser General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public
+  License along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <sys/types.h>
+#include <errno.h>
+
+#ifdef HAVE_PWD_H
+#include <pwd.h>
+#endif
+
+#ifdef HAVE_GRP_H
+#include <grp.h>
+#endif
+
+#include <pulse/xmalloc.h>
+#include <pulsecore/macro.h>
+
+#include "usergroup.h"
+
+#ifdef HAVE_GRP_H
+
+/* Returns a suitable starting size for a getgrnam_r() or getgrgid_r() buffer,
+   plus the size of a struct group.
+ */
+static size_t starting_getgr_buflen(void) {
+    size_t full_size;
+    long n;
+#ifdef _SC_GETGR_R_SIZE_MAX
+    n = sysconf(_SC_GETGR_R_SIZE_MAX);
+#else
+    n = -1;
+#endif
+    if (n <= 0)
+        n = 512;
+
+    full_size = (size_t) n + sizeof(struct group);
+
+    if (full_size < (size_t) n) /* check for integer overflow */
+        return (size_t) n;
+
+    return full_size;
+}
+
+/* Returns a suitable starting size for a getpwnam_r() or getpwuid_r() buffer,
+   plus the size of a struct passwd.
+ */
+static size_t starting_getpw_buflen(void) {
+    long n;
+    size_t full_size;
+
+#ifdef _SC_GETPW_R_SIZE_MAX
+    n = sysconf(_SC_GETPW_R_SIZE_MAX);
+#else
+    n = -1;
+#endif
+    if (n <= 0)
+        n = 512;
+
+    full_size = (size_t) n + sizeof(struct passwd);
+
+    if (full_size < (size_t) n) /* check for integer overflow */
+        return (size_t) n;
+
+    return full_size;
+}
+
+/* Given a memory allocation (*bufptr) and its length (*buflenptr),
+   double the size of the allocation, updating the given buffer and length
+   arguments. This function should be used in conjunction with the pa_*alloc
+   and pa_xfree functions.
+
+   Unlike realloc(), this function does *not* retain the original buffer's
+   contents.
+
+   Returns 0 on success, nonzero on error. The error cause is indicated by
+   errno.
+ */
+static int expand_buffer_trashcontents(void **bufptr, size_t *buflenptr) {
+    size_t newlen;
+
+    if (!bufptr || !*bufptr || !buflenptr) {
+        errno = EINVAL;
+        return -1;
+    }
+
+    newlen = *buflenptr * 2;
+
+    if (newlen < *buflenptr) {
+        errno = EOVERFLOW;
+        return -1;
+    }
+
+    /* Don't bother retaining memory contents; free & alloc anew */
+    pa_xfree(*bufptr);
+
+    *bufptr = pa_xmalloc(newlen);
+    *buflenptr = newlen;
+
+    return 0;
+}
+
+#ifdef HAVE_GETGRGID_R
+/* Thread-safe getgrgid() replacement.
+   Returned value should be freed using pa_getgrgid_free() when the caller is
+   finished with the returned group data.
+
+   API is the same as getgrgid(), errors are indicated by a NULL return;
+   consult errno for the error cause (zero it before calling).
+ */
+struct group *pa_getgrgid_malloc(gid_t gid) {
+    size_t buflen, getgr_buflen;
+    int err;
+    void *buf;
+    void *getgr_buf;
+    struct group *result = NULL;
+
+    buflen = starting_getgr_buflen();
+    buf = pa_xmalloc(buflen);
+
+    getgr_buflen = buflen - sizeof(struct group);
+    getgr_buf = (char *)buf + sizeof(struct group);
+
+    while ((err = getgrgid_r(gid, (struct group *)buf, getgr_buf,
+                    getgr_buflen, &result)) == ERANGE)
+    {
+        if (expand_buffer_trashcontents(&buf, &buflen))
+            break;
+
+        getgr_buflen = buflen - sizeof(struct group);
+        getgr_buf = (char *)buf + sizeof(struct group);
+    }
+
+    if (err || !result) {
+        result = NULL;
+        if (buf) {
+            pa_xfree(buf);
+            buf = NULL;
+        }
+    }
+
+    pa_assert(result == buf || result == NULL);
+
+    return result;
+}
+
+void pa_getgrgid_free(struct group *grp) {
+    pa_xfree(grp);
+}
+
+#else /* !HAVE_GETGRGID_R */
+
+struct group *pa_getgrgid_malloc(gid_t gid) {
+    return getgrgid(gid);
+}
+
+void pa_getgrgid_free(struct group *grp) {
+    /* nothing */
+    return;
+}
+
+#endif /* !HAVE_GETGRGID_R */
+
+#ifdef HAVE_GETGRNAM_R
+/* Thread-safe getgrnam() function.
+   Returned value should be freed using pa_getgrnam_free() when the caller is
+   finished with the returned group data.
+
+   API is the same as getgrnam(), errors are indicated by a NULL return;
+   consult errno for the error cause (zero it before calling).
+ */
+struct group *pa_getgrnam_malloc(const char *name) {
+    size_t buflen, getgr_buflen;
+    int err;
+    void *buf;
+    void *getgr_buf;
+    struct group *result = NULL;
+
+    buflen = starting_getgr_buflen();
+    buf = pa_xmalloc(buflen);
+
+    getgr_buflen = buflen - sizeof(struct group);
+    getgr_buf = (char *)buf + sizeof(struct group);
+
+    while ((err = getgrnam_r(name, (struct group *)buf, getgr_buf,
+                    getgr_buflen, &result)) == ERANGE)
+    {
+        if (expand_buffer_trashcontents(&buf, &buflen))
+            break;
+
+        getgr_buflen = buflen - sizeof(struct group);
+        getgr_buf = (char *)buf + sizeof(struct group);
+    }
+
+    if (err || !result) {
+        result = NULL;
+        if (buf) {
+            pa_xfree(buf);
+            buf = NULL;
+        }
+    }
+
+    pa_assert(result == buf || result == NULL);
+
+    return result;
+}
+
+void pa_getgrnam_free(struct group *group) {
+    pa_xfree(group);
+}
+
+#else /* !HAVE_GETGRNAM_R */
+
+struct group *pa_getgrnam_malloc(const char *name) {
+    return getgrnam(name);
+}
+
+void pa_getgrnam_free(struct group *group) {
+    /* nothing */
+    return;
+}
+
+#endif /* HAVE_GETGRNAM_R */
+
+#endif /* HAVE_GRP_H */
+
+#ifdef HAVE_PWD_H
+
+#ifdef HAVE_GETPWNAM_R
+/* Thread-safe getpwnam() function.
+   Returned value should be freed using pa_getpwnam_free() when the caller is
+   finished with the returned passwd data.
+
+   API is the same as getpwnam(), errors are indicated by a NULL return;
+   consult errno for the error cause (zero it before calling).
+ */
+struct passwd *pa_getpwnam_malloc(const char *name) {
+    size_t buflen, getpw_buflen;
+    int err;
+    void *buf;
+    void *getpw_buf;
+    struct passwd *result = NULL;
+
+    buflen = starting_getpw_buflen();
+    buf = pa_xmalloc(buflen);
+
+    getpw_buflen = buflen - sizeof(struct passwd);
+    getpw_buf = (char *)buf + sizeof(struct passwd);
+
+    while ((err = getpwnam_r(name, (struct passwd *)buf, getpw_buf,
+                    getpw_buflen, &result)) == ERANGE)
+    {
+        if (expand_buffer_trashcontents(&buf, &buflen))
+            break;
+
+        getpw_buflen = buflen - sizeof(struct passwd);
+        getpw_buf = (char *)buf + sizeof(struct passwd);
+    }
+
+    if (err || !result) {
+        result = NULL;
+        if (buf) {
+            pa_xfree(buf);
+            buf = NULL;
+        }
+    }
+
+    pa_assert(result == buf || result == NULL);
+
+    return result;
+}
+
+void pa_getpwnam_free(struct passwd *passwd) {
+    pa_xfree(passwd);
+}
+
+#else /* !HAVE_GETPWNAM_R */
+
+struct passwd *pa_getpwnam_malloc(const char *name) {
+    return getpwnam(name);
+}
+
+void pa_getpwnam_free(struct passwd *passwd) {
+    /* nothing */
+    return;
+}
+
+#endif /* !HAVE_GETPWNAM_R */
+
+#ifdef HAVE_GETPWUID_R
+/* Thread-safe getpwuid() function.
+   Returned value should be freed using pa_getpwuid_free() when the caller is
+   finished with the returned group data.
+
+   API is the same as getpwuid(), errors are indicated by a NULL return;
+   consult errno for the error cause (zero it before calling).
+ */
+struct passwd *pa_getpwuid_malloc(uid_t uid) {
+    size_t buflen, getpw_buflen;
+    int err;
+    void *buf;
+    void *getpw_buf;
+    struct passwd *result = NULL;
+
+    buflen = starting_getpw_buflen();
+    buf = pa_xmalloc(buflen);
+
+    getpw_buflen = buflen - sizeof(struct passwd);
+    getpw_buf = (char *)buf + sizeof(struct passwd);
+
+    while ((err = getpwuid_r(uid, (struct passwd *)buf, getpw_buf,
+                    getpw_buflen, &result)) == ERANGE)
+    {
+        if (expand_buffer_trashcontents(&buf, &buflen))
+            break;
+
+        getpw_buflen = buflen - sizeof(struct passwd);
+        getpw_buf = (char *)buf + sizeof(struct passwd);
+    }
+
+    if (err || !result) {
+        result = NULL;
+        if (buf) {
+            pa_xfree(buf);
+            buf = NULL;
+        }
+    }
+
+    pa_assert(result == buf || result == NULL);
+
+    return result;
+}
+
+void pa_getpwuid_free(struct passwd *passwd) {
+    pa_xfree(passwd);
+}
+
+#else /* !HAVE_GETPWUID_R */
+
+struct passwd *pa_getpwuid_malloc(uid_t uid) {
+    return getpwuid(uid);
+}
+
+void pa_getpwuid_free(struct passwd *passwd) {
+    /* nothing */
+    return;
+}
+
+#endif /* !HAVE_GETPWUID_R */
+
+#endif /* HAVE_PWD_H */
diff --git a/src/pulsecore/usergroup.h b/src/pulsecore/usergroup.h
new file mode 100644
index 00000000..1c091638
--- /dev/null
+++ b/src/pulsecore/usergroup.h
@@ -0,0 +1,51 @@
+#ifndef foousergrouphfoo
+#define foousergrouphfoo
+
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2009 Ted Percival
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as
+  published by the Free Software Foundation; either version 2.1 of the
+  License, or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  Lesser General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public
+  License along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#include <sys/types.h>
+
+#ifndef PACKAGE
+#error "Please include config.h before including this file!"
+#endif
+
+#ifdef HAVE_GRP_H
+
+struct group *pa_getgrgid_malloc(gid_t gid);
+void pa_getgrgid_free(struct group *grp);
+
+struct group *pa_getgrnam_malloc(const char *name);
+void pa_getgrnam_free(struct group *group);
+
+#endif /* HAVE_GRP_H */
+
+#ifdef HAVE_PWD_H
+
+struct passwd *pa_getpwuid_malloc(uid_t uid);
+void pa_getpwuid_free(struct passwd *passwd);
+
+struct passwd *pa_getpwnam_malloc(const char *name);
+void pa_getpwnam_free(struct passwd *passwd);
+
+#endif /* HAVE_PWD_H */
+
+#endif /* foousergrouphfoo */
diff --git a/src/tests/envelope-test.c b/src/tests/envelope-test.c
index 3af3044e..9382040b 100644
--- a/src/tests/envelope-test.c
+++ b/src/tests/envelope-test.c
@@ -34,8 +34,6 @@
 #include <pulsecore/memblock.h>
 #include <pulsecore/sample-util.h>
 
-#include <liboil/liboil.h>
-
 const pa_envelope_def ramp_down = {
     .n_points = 2,
     .points_x = { 100*PA_USEC_PER_MSEC, 300*PA_USEC_PER_MSEC },
@@ -202,7 +200,6 @@ int main(int argc, char *argv[]) {
         .values = { PA_VOLUME_NORM, PA_VOLUME_NORM/2 }
     };
 
-    oil_init();
     pa_log_set_level(PA_LOG_DEBUG);
 
     pa_assert_se(pool = pa_mempool_new(FALSE, 0));
diff --git a/src/tests/mix-test.c b/src/tests/mix-test.c
index f9f76da3..457c4acd 100644
--- a/src/tests/mix-test.c
+++ b/src/tests/mix-test.c
@@ -32,8 +32,6 @@
 #include <pulsecore/memblock.h>
 #include <pulsecore/sample-util.h>
 
-#include <liboil/liboil.h>
-
 static float swap_float(float a) {
     uint32_t *b = (uint32_t*) &a;
     *b = PA_UINT32_SWAP(*b);
@@ -211,7 +209,6 @@ int main(int argc, char *argv[]) {
     pa_sample_spec a;
     pa_cvolume v;
 
-    oil_init();
     pa_log_set_level(PA_LOG_DEBUG);
 
     pa_assert_se(pool = pa_mempool_new(FALSE, 0));
diff --git a/src/tests/remix-test.c b/src/tests/remix-test.c
index 9d110d6b..4990bf93 100644
--- a/src/tests/remix-test.c
+++ b/src/tests/remix-test.c
@@ -32,8 +32,6 @@
 #include <pulsecore/memblock.h>
 #include <pulsecore/sample-util.h>
 
-#include <liboil/liboil.h>
-
 int main(int argc, char *argv[]) {
 
     static const pa_channel_map maps[] = {
@@ -55,7 +53,6 @@ int main(int argc, char *argv[]) {
     unsigned i, j;
     pa_mempool *pool;
 
-    oil_init();
     pa_log_set_level(PA_LOG_DEBUG);
 
     pa_assert_se(pool = pa_mempool_new(FALSE, 0));
diff --git a/src/tests/resampler-test.c b/src/tests/resampler-test.c
index 7236265a..82198b5e 100644
--- a/src/tests/resampler-test.c
+++ b/src/tests/resampler-test.c
@@ -32,8 +32,6 @@
 #include <pulsecore/memblock.h>
 #include <pulsecore/sample-util.h>
 
-#include <liboil/liboil.h>
-
 static void dump_block(const pa_sample_spec *ss, const pa_memchunk *chunk) {
     void *d;
     unsigned i;
@@ -248,7 +246,6 @@ int main(int argc, char *argv[]) {
     pa_sample_spec a, b;
     pa_cvolume v;
 
-    oil_init();
     pa_log_set_level(PA_LOG_DEBUG);
 
     pa_assert_se(pool = pa_mempool_new(FALSE, 0));
diff --git a/src/tests/usergroup-test.c b/src/tests/usergroup-test.c
new file mode 100644
index 00000000..a48b016d
--- /dev/null
+++ b/src/tests/usergroup-test.c
@@ -0,0 +1,161 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2009 Ted Percival
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as
+  published by the Free Software Foundation; either version 2.1 of the
+  License, or (at your option) any later version.
+
+  PulseAudio is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  Lesser General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public
+  License along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#include <config.h>
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/types.h>
+#include <pwd.h>
+#include <grp.h>
+#include <errno.h>
+
+#include <pulsecore/usergroup.h>
+
+static int load_reference_structs(struct group **gr, struct passwd **pw) {
+    setpwent();
+    *pw = getpwent();
+    endpwent();
+
+    setgrent();
+    *gr = getgrent();
+    endgrent();
+
+    return (*gr && *pw) ? 0 : 1;
+}
+
+static int compare_group(const struct group *a, const struct group *b) {
+    char **amem, **bmem;
+
+    if (strcmp(a->gr_name, b->gr_name)) {
+        fprintf(stderr, "Group name mismatch: [%s] [%s]\n",
+                a->gr_name, b->gr_name);
+        return 1;
+    }
+
+    if (strcmp(a->gr_passwd, b->gr_passwd)) {
+        fprintf(stderr, "Group password mismatch: [%s] [%s]\n",
+                a->gr_passwd, b->gr_passwd);
+        return 1;
+    }
+
+    if (a->gr_gid != b->gr_gid) {
+        fprintf(stderr, "Gid mismatch: [%lu] [%lu]\n",
+                (unsigned long) a->gr_gid, (unsigned long) b->gr_gid);
+        return 1;
+    }
+
+    /* XXX: Assuming the group ordering is identical. */
+    for (amem = a->gr_mem, bmem = b->gr_mem; *amem && *bmem; ++amem, ++bmem) {
+        if (strcmp(*amem, *bmem)) {
+            fprintf(stderr, "Group member mismatch: [%s] [%s]\n",
+                    *amem, *bmem);
+            return 1;
+        }
+    }
+
+    if (*amem || *bmem) {
+        fprintf(stderr, "Mismatched group count\n");
+        return 1;
+    }
+
+    return 0;
+}
+
+static int compare_passwd(const struct passwd *a, const struct passwd *b) {
+    if (strcmp(a->pw_name, b->pw_name)) {
+        fprintf(stderr, "pw_name mismatch: [%s] [%s]\n", a->pw_name, b->pw_name);
+        return 1;
+    }
+
+    if (strcmp(a->pw_passwd, b->pw_passwd)) {
+        fprintf(stderr, "pw_passwd mismatch: [%s] [%s]\n", a->pw_passwd, b->pw_passwd);
+        return 1;
+    }
+
+    if (a->pw_uid != b->pw_uid) {
+        fprintf(stderr, "pw_uid mismatch: [%lu] [%lu]\n",
+		(unsigned long) a->pw_uid, (unsigned long) b->pw_uid);
+        return 1;
+    }
+
+    if (a->pw_gid != b->pw_gid) {
+        fprintf(stderr, "pw_gid mismatch: [%lu] [%lu]\n",
+		(unsigned long) a->pw_gid, (unsigned long) b->pw_gid);
+        return 1;
+    }
+
+    if (strcmp(a->pw_gecos, b->pw_gecos)) {
+        fprintf(stderr, "pw_gecos mismatch: [%s] [%s]\n", a->pw_gecos, b->pw_gecos);
+        return 1;
+    }
+
+    if (strcmp(a->pw_dir, b->pw_dir)) {
+        fprintf(stderr, "pw_dir mismatch: [%s] [%s]\n", a->pw_dir, b->pw_dir);
+        return 1;
+    }
+
+    if (strcmp(a->pw_shell, b->pw_shell)) {
+        fprintf(stderr, "pw_shell mismatch: [%s] [%s]\n", a->pw_shell, b->pw_shell);
+        return 1;
+    }
+
+    return 0;
+}
+
+int main(int argc, char *argv[]) {
+    struct group *gr;
+    struct passwd *pw;
+    int err;
+    struct group *reference_group = NULL;
+    struct passwd *reference_passwd = NULL;
+
+    err = load_reference_structs(&reference_group, &reference_passwd);
+    if (err)
+        return 77;
+
+    errno = 0;
+    gr = pa_getgrgid_malloc(reference_group->gr_gid);
+    if (compare_group(reference_group, gr))
+        return 1;
+    pa_getgrgid_free(gr);
+
+    errno = 0;
+    gr = pa_getgrnam_malloc(reference_group->gr_name);
+    if (compare_group(reference_group, gr))
+        return 1;
+    pa_getgrnam_free(gr);
+
+    errno = 0;
+    pw = pa_getpwuid_malloc(reference_passwd->pw_uid);
+    if (compare_passwd(reference_passwd, pw))
+        return 1;
+    pa_getpwuid_free(pw);
+
+    errno = 0;
+    pw = pa_getpwnam_malloc(reference_passwd->pw_name);
+    if (compare_passwd(reference_passwd, pw))
+        return 1;
+    pa_getpwnam_free(pw);
+
+    return 0;
+}