[3DS] Enable threading

Makefile.common

@@ -686,7 +686,8 @@ endif
 
 ifeq ($(TARGET), retroarch_3ds)
    OBJ += audio/drivers/ctr_csnd_audio.o \
-          audio/drivers/ctr_dsp_audio.o
+          audio/drivers/ctr_dsp_audio.o \
+          audio/drivers/ctr_dsp_thread_audio.o \
 endif
 
 ifeq ($(HAVE_ALSA), 1)

Makefile.ctr

@@ -56,7 +56,8 @@ ifeq ($(GRIFFIN_BUILD), 1)
 	DEFINES += -DHAVE_PATCH -DHAVE_RWAV
 	DEFINES += -DHAVE_SCREENSHOTS
 	DEFINES += -DHAVE_REWIND
-	#DEFINES += -DHAVE_SOCKET_LEGACY -DHAVE_THREADS
+	DEFINES += -DHAVE_THREADS
+	#DEFINES += -DHAVE_SOCKET_LEGACY
 	#-DHAVE_SSL -DHAVE_BUILTINMBEDTLS -DMBEDTLS_SSL_DEBUG_ALL
 	#ssl is currently incompatible with griffin due to use of the "static" flag on repeating functions that will conflict when included in one file
 else
@@ -84,7 +85,7 @@ else
 	#HAVE_NETWORKING = 1
 	#HAVE_CHEEVOS = 1
 	#HAVE_SOCKET_LEGACY = 1
-	#HAVE_THREADS = 1
+	HAVE_THREADS = 1
 	#HAVE_SSL = 1
 	#HAVE_BUILTINMBEDTLS = 1
 

audio/drivers/ctr_dsp_thread_audio.c

@@ -0,0 +1,339 @@
+/*  RetroArch - A frontend for libretro.
+ *  Copyright (C) 2020 - Justin Weiss
+ *
+ *  RetroArch is free software: you can redistribute it and/or modify it under the terms
+ *  of the GNU General Public License as published by the Free Software Found-
+ *  ation, either version 3 of the License, or (at your option) any later version.
+ *
+ *  RetroArch 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 General Public License along with RetroArch.
+ *  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <3ds.h>
+#include <string.h>
+#include <malloc.h>
+#include <queues/fifo_queue.h>
+#include <rthreads/rthreads.h>
+
+#include "../../retroarch.h"
+#include "../../ctr/ctr_debug.h"
+
+typedef struct
+{
+   fifo_buffer_t* fifo;
+   size_t fifo_size;
+
+   slock_t* fifo_lock;
+   scond_t* fifo_avail;
+   scond_t* fifo_done;
+   sthread_t* thread;
+
+   volatile bool running;
+
+   bool nonblocking;
+   bool playing;
+   retro_time_t frame_time;
+   int channel;
+   ndspWaveBuf dsp_buf;
+
+   uint32_t pos;
+} ctr_dsp_thread_audio_t;
+
+// PCM16 stereo
+#define DSP_BYTES_TO_SAMPLES(bytes) (bytes / (2 * sizeof(uint16_t)))
+#define DSP_SAMPLES_TO_BYTES(samples) (samples * 2 * sizeof(uint16_t))
+
+static void ctr_dsp_audio_loop(void* data)
+{
+   uint32_t pos, buf_pos;
+   ctr_dsp_thread_audio_t *ctr  = (ctr_dsp_thread_audio_t*)data;
+
+   if (!ctr)
+      return;
+
+   while (1)
+   {
+      size_t buf_avail, avail, to_write;
+      slock_lock(ctr->fifo_lock);
+
+      do {
+         avail = FIFO_READ_AVAIL(ctr->fifo);
+
+         if (!avail) {
+            scond_wait(ctr->fifo_avail, ctr->fifo_lock);
+         }
+      } while (!avail && ctr->running);
+
+      slock_unlock(ctr->fifo_lock);
+
+      if (!ctr->running)
+         break;
+
+      pos = ctr->pos;
+      buf_pos = DSP_SAMPLES_TO_BYTES(ndspChnGetSamplePos(ctr->channel));
+
+      buf_avail = buf_pos >= pos ? buf_pos - pos : ctr->fifo_size - pos;
+      to_write = MIN(avail, buf_avail);
+
+      slock_lock(ctr->fifo_lock);
+
+      if (to_write > 0) {
+         fifo_read(ctr->fifo, ctr->dsp_buf.data_pcm8 + pos, to_write);
+         DSP_FlushDataCache(ctr->dsp_buf.data_pcm8 + pos, to_write);
+         scond_signal(ctr->fifo_done);
+      }
+
+      slock_unlock(ctr->fifo_lock);
+
+      if (buf_pos == pos) {
+         svcSleepThread(100000);
+      }
+
+      ctr->pos = (pos + to_write) % ctr->fifo_size;
+   }
+}
+
+static void ctr_dsp_thread_audio_free(void *data);
+
+static void *ctr_dsp_thread_audio_init(const char *device, unsigned rate, unsigned latency,
+      unsigned block_frames,
+      unsigned *new_rate)
+{
+   ctr_dsp_thread_audio_t *ctr = NULL;
+
+   (void)device;
+   (void)rate;
+
+   if (ndspInit() < 0)
+      return NULL;
+
+   ctr = (ctr_dsp_thread_audio_t*)calloc(1, sizeof(ctr_dsp_thread_audio_t));
+
+   if (!ctr)
+      return NULL;
+
+   *new_rate    = 32728;
+
+   ctr->running = true;
+   ctr->channel = 0;
+
+   ndspSetOutputMode(NDSP_OUTPUT_STEREO);
+   ndspSetClippingMode(NDSP_CLIP_SOFT); /* ?? */
+   ndspSetOutputCount(2);
+   ndspChnReset(ctr->channel);
+   ndspChnSetFormat(ctr->channel, NDSP_FORMAT_STEREO_PCM16);
+   ndspChnSetInterp(ctr->channel, NDSP_INTERP_NONE);
+   ndspChnSetRate(ctr->channel, 32728.0f);
+   ndspChnWaveBufClear(ctr->channel);
+
+   ctr->fifo_size = DSP_SAMPLES_TO_BYTES((*new_rate * MAX(latency, 8)) / 1000);
+
+   ctr->dsp_buf.data_pcm16 = linearAlloc(ctr->fifo_size);
+   memset(ctr->dsp_buf.data_pcm16, 0, ctr->fifo_size);
+   DSP_FlushDataCache(ctr->dsp_buf.data_pcm16, ctr->fifo_size);
+
+   ctr->dsp_buf.looping = true;
+   ctr->dsp_buf.nsamples = DSP_BYTES_TO_SAMPLES(ctr->fifo_size);
+   ndspChnWaveBufAdd(ctr->channel, &ctr->dsp_buf);
+
+   ctr->fifo = fifo_new(ctr->fifo_size);
+
+   if (!(ctr->fifo_lock = slock_new()) ||
+       !(ctr->fifo_avail = scond_new()) ||
+       !(ctr->fifo_done = scond_new()) ||
+       !(ctr->thread = sthread_create(ctr_dsp_audio_loop, ctr)))
+   {
+      RARCH_LOG("[Audio]: thread creation failed.\n");
+      ctr->running = false;
+      ctr_dsp_thread_audio_free(ctr);
+      return NULL;
+   }
+
+   ctr->pos = 0;
+   ctr->playing = true;
+
+   ctr->frame_time = (retro_time_t)roundf(1000000 * 4481134.0 / (*new_rate * 8192.0));
+
+   ndspSetMasterVol(1.0);
+
+   return ctr;
+}
+
+static void ctr_dsp_thread_audio_free(void *data)
+{
+   ctr_dsp_thread_audio_t* ctr = (ctr_dsp_thread_audio_t*)data;
+   if (!ctr)
+      return;
+
+   if (ctr->running)
+   {
+      ctr->running = false;
+      scond_signal(ctr->fifo_avail);
+   }
+
+   if (ctr->thread)
+      sthread_join(ctr->thread);
+
+   scond_free(ctr->fifo_avail);
+   scond_free(ctr->fifo_done);
+   slock_free(ctr->fifo_lock);
+
+   if (ctr->fifo)
+   {
+      fifo_free(ctr->fifo);
+      ctr->fifo = NULL;
+   }
+
+   ndspChnWaveBufClear(ctr->channel);
+
+   linearFree(ctr->dsp_buf.data_pcm16);
+
+   free(ctr);
+   ndspExit();
+   ctr = NULL;
+}
+
+static ssize_t ctr_dsp_thread_audio_write(void *data, const void *buf, size_t size)
+{
+   size_t avail, written;
+   ctr_dsp_thread_audio_t * ctr = (ctr_dsp_thread_audio_t*)data;
+
+   if (!ctr || !ctr->running)
+      return 0;
+
+   if (ctr->nonblocking)
+   {
+      slock_lock(ctr->fifo_lock);
+      avail = FIFO_WRITE_AVAIL(ctr->fifo);
+      written = MIN(avail, size);
+      if (written > 0)
+      {
+         fifo_write(ctr->fifo, buf, written);
+         scond_signal(ctr->fifo_avail);
+      }
+      slock_unlock(ctr->fifo_lock);
+   }
+   else
+   {
+      written = 0;
+      while (written < size && ctr->running)
+      {
+         slock_lock(ctr->fifo_lock);
+         avail = FIFO_WRITE_AVAIL(ctr->fifo);
+         if (avail == 0)
+         {
+            if (ctr->running)
+            {
+               /* Wait a maximum of one frame, skip the write if the thread is still busy */
+               if (!scond_wait_timeout(ctr->fifo_done, ctr->fifo_lock, ctr->frame_time)) {
+                  slock_unlock(ctr->fifo_lock);
+                  break;
+               }
+            }
+            slock_unlock(ctr->fifo_lock);
+         }
+         else
+         {
+            size_t write_amt = MIN(size - written, avail);
+            fifo_write(ctr->fifo, (const char*)buf + written, write_amt);
+            scond_signal(ctr->fifo_avail);
+            slock_unlock(ctr->fifo_lock);
+            written += write_amt;
+         }
+      }
+   }
+
+   return written;
+}
+
+static bool ctr_dsp_thread_audio_stop(void *data)
+{
+   ctr_dsp_thread_audio_t* ctr = (ctr_dsp_thread_audio_t*)data;
+
+   if (!ctr)
+     return false;
+
+   ndspSetMasterVol(0.0);
+   ctr->playing = false;
+
+   return true;
+}
+
+static bool ctr_dsp_thread_audio_alive(void *data)
+{
+   ctr_dsp_thread_audio_t* ctr = (ctr_dsp_thread_audio_t*)data;
+
+   if (!ctr)
+     return false;
+
+   return ctr->playing;
+}
+
+static bool ctr_dsp_thread_audio_start(void *data, bool is_shutdown)
+{
+   ctr_dsp_thread_audio_t* ctr = (ctr_dsp_thread_audio_t*)data;
+
+   if (!ctr)
+     return false;
+
+   /* Prevents restarting audio when the menu
+    * is toggled off on shutdown */
+   if (is_shutdown)
+      return true;
+
+   ndspSetMasterVol(1.0);
+   ctr->playing = true;
+
+   return true;
+}
+
+static void ctr_dsp_thread_audio_set_nonblock_state(void *data, bool state)
+{
+   ctr_dsp_thread_audio_t* ctr = (ctr_dsp_thread_audio_t*)data;
+   if (ctr)
+      ctr->nonblocking = state;
+}
+
+static bool ctr_dsp_thread_audio_use_float(void *data)
+{
+   (void)data;
+   return false;
+}
+
+static size_t ctr_dsp_thread_audio_write_avail(void *data)
+{
+   size_t val;
+   ctr_dsp_thread_audio_t* ctr = (ctr_dsp_thread_audio_t*)data;
+
+   slock_lock(ctr->fifo_lock);
+   val = FIFO_WRITE_AVAIL(ctr->fifo);
+   slock_unlock(ctr->fifo_lock);
+
+   return val;
+}
+
+static size_t ctr_dsp_thread_audio_buffer_size(void *data)
+{
+   ctr_dsp_thread_audio_t* ctr = (ctr_dsp_thread_audio_t*)data;
+   return ctr->fifo_size;
+}
+
+audio_driver_t audio_ctr_dsp_thread = {
+   ctr_dsp_thread_audio_init,
+   ctr_dsp_thread_audio_write,
+   ctr_dsp_thread_audio_stop,
+   ctr_dsp_thread_audio_start,
+   ctr_dsp_thread_audio_alive,
+   ctr_dsp_thread_audio_set_nonblock_state,
+   ctr_dsp_thread_audio_free,
+   ctr_dsp_thread_audio_use_float,
+   "dsp_thread",
+   NULL,
+   NULL,
+   ctr_dsp_thread_audio_write_avail,
+   ctr_dsp_thread_audio_buffer_size
+};

frontend/drivers/platform_ctr.c

@@ -482,8 +482,10 @@ static void frontend_ctr_init(void* data)
    if (csndInit() != 0)
       audio_ctr_csnd = audio_null;
    ctr_check_dspfirm();
-   if (ndspInit() != 0)
+   if (ndspInit() != 0) {
       audio_ctr_dsp = audio_null;
+      audio_ctr_dsp_thread = audio_null;
+   }
    cfguInit();
    ptmuInit();
    mcuHwcInit();

gfx/drivers/ctr_gfx.c

@@ -567,12 +567,6 @@ static bool ctr_frame(void* data, const void* frame,
       return true;
    }
 
-   if(!aptMainLoop())
-   {
-      command_event(CMD_EVENT_QUIT, NULL);
-      return true;
-   }
-
    state_tmp = hidKeysDown();
    hidTouchRead(&state_tmp_touch);
    if((state_tmp & KEY_TOUCH) && (state_tmp_touch.py < 120))

gfx/video_thread_wrapper.c

@@ -822,7 +822,11 @@ static bool video_thread_init(thread_video_t *thr,
    max_size                  = info.input_scale * RARCH_SCALE_BASE;
    max_size                 *= max_size;
    max_size                 *= info.rgb32 ? sizeof(uint32_t) : sizeof(uint16_t);
+#ifdef _3DS
+   thr->frame.buffer         = linearMemAlign(max_size, 0x80);
+#else
    thr->frame.buffer         = (uint8_t*)malloc(max_size);
+#endif
 
    if (!thr->frame.buffer)
       return false;
@@ -937,7 +941,11 @@ static void video_thread_free(void *data)
 #if defined(HAVE_MENU)
    free(thr->texture.frame);
 #endif
+#ifdef _3DS
+   linearFree(thr->frame.buffer);
+#else
    free(thr->frame.buffer);
+#endif
    slock_free(thr->frame.lock);
    slock_free(thr->lock);
    scond_free(thr->cond_cmd);

griffin/griffin.c

@@ -905,6 +905,7 @@ AUDIO
 #elif defined(_3DS)
 #include "../audio/drivers/ctr_csnd_audio.c"
 #include "../audio/drivers/ctr_dsp_audio.c"
+#include "../audio/drivers/ctr_dsp_thread_audio.c"
 #endif
 
 #ifdef HAVE_XAUDIO

libretro-common/rthreads/ctr_pthread.h

@@ -30,7 +30,7 @@
 #include <errno.h>
 #include <retro_inline.h>
 
-#define STACKSIZE (4 * 1024)
+#define STACKSIZE (32 * 1024)
 
 #ifndef PTHREAD_SCOPE_PROCESS
 /* An earlier version of devkitARM does not define the pthread types. Can remove in r54+. */
@@ -39,10 +39,16 @@ typedef Thread     pthread_t;
 typedef LightLock  pthread_mutex_t;
 typedef void*      pthread_mutexattr_t;
 typedef int        pthread_attr_t;
-typedef LightEvent pthread_cond_t;
+typedef uint32_t   pthread_cond_t;
 typedef int        pthread_condattr_t;
 #endif
 
+typedef struct {
+   uint32_t semaphore;
+   LightLock lock;
+   uint32_t waiting;
+} cond_t;
+
 /* libctru threads return void but pthreads return void pointer */
 static bool mutex_inited = false;
 static LightLock safe_double_thread_launch;
@@ -60,11 +66,18 @@ static INLINE int pthread_create(pthread_t *thread,
 {
    s32 prio = 0;
    Thread new_ctr_thread;
+   int procnum = -2; // use default cpu
+   bool isNew3DS;
+
+   APT_CheckNew3DS(&isNew3DS);
+
+   if (isNew3DS)
+      procnum = 2;
 
    if (!mutex_inited)
    {
-	   LightLock_Init(&safe_double_thread_launch);
+      LightLock_Init(&safe_double_thread_launch);
-	   mutex_inited = true;
+      mutex_inited = true;
    }
 
    /*Must wait if attempting to launch 2 threads at once to prevent corruption of function pointer*/
@@ -73,12 +86,12 @@ static INLINE int pthread_create(pthread_t *thread,
    svcGetThreadPriority(&prio, CUR_THREAD_HANDLE);
 
    start_routine_jump = start_routine;
-   new_ctr_thread     = threadCreate(ctr_thread_launcher, arg, STACKSIZE, prio - 1, -1/*No affinity, use any CPU*/, false);
+   new_ctr_thread     = threadCreate(ctr_thread_launcher, arg, STACKSIZE, prio - 1, procnum, false);
 
    if (!new_ctr_thread)
    {
-	   LightLock_Unlock(&safe_double_thread_launch);
+      LightLock_Unlock(&safe_double_thread_launch);
-	   return EAGAIN;
+      return EAGAIN;
    }
 
    *thread = (pthread_t)new_ctr_thread;
@@ -105,7 +118,8 @@ static INLINE int pthread_mutex_destroy(pthread_mutex_t *mutex)
 
 static INLINE int pthread_mutex_lock(pthread_mutex_t *mutex)
 {
-   return LightLock_TryLock((LightLock *)mutex);
+   LightLock_Lock((LightLock *)mutex);
+   return 0;
 }
 
 static INLINE int pthread_mutex_unlock(pthread_mutex_t *mutex)
@@ -119,6 +133,8 @@ static INLINE void pthread_exit(void *retval)
    /*Yes the pointer to int cast is not ideal*/
    /*threadExit((int)retval);*/
    (void)retval;
+
+   threadExit(0);
 }
 
 static INLINE int pthread_detach(pthread_t thread)
@@ -131,7 +147,12 @@ static INLINE int pthread_detach(pthread_t thread)
 static INLINE int pthread_join(pthread_t thread, void **retval)
 {
    /*retval is ignored*/
-   return threadJoin((Thread)thread, U64_MAX);
+   if(threadJoin((Thread)thread, INT64_MAX))
+      return -1;
+
+   threadFree((Thread)thread);
+
+   return 0;
 }
 
 static INLINE int pthread_mutex_trylock(pthread_mutex_t *mutex)
@@ -142,51 +163,109 @@ static INLINE int pthread_mutex_trylock(pthread_mutex_t *mutex)
 static INLINE int pthread_cond_wait(pthread_cond_t *cond,
       pthread_mutex_t *mutex)
 {
-   LightEvent_Wait((LightEvent *)cond);
+   cond_t *cond_data = (cond_t *)*cond;
+   LightLock_Lock(&cond_data->lock);
+   cond_data->waiting++;
+   LightLock_Unlock(mutex);
+   LightLock_Unlock(&cond_data->lock);
+   svcWaitSynchronization(cond_data->semaphore, INT64_MAX);
+   LightLock_Lock(mutex);
    return 0;
 }
 
 static INLINE int pthread_cond_timedwait(pthread_cond_t *cond,
       pthread_mutex_t *mutex, const struct timespec *abstime)
 {
-   for (;;)
+   struct timespec now = {0};
-   {
+   /* Missing clock_gettime*/
-       struct timespec now = {0};
+   struct timeval tm;
-	    /* Missing clock_gettime*/
+   int retval = 0;
-       struct timeval tm;
 
-       gettimeofday(&tm, NULL);
+   cond_t *cond_data = (cond_t *)*cond;
-       now.tv_sec = tm.tv_sec;
+   LightLock_Lock(&cond_data->lock);
-       now.tv_nsec = tm.tv_usec * 1000;
+   cond_data->waiting++;
-       if (LightEvent_TryWait((LightEvent *)cond) != 0 || now.tv_sec > abstime->tv_sec || (now.tv_sec == abstime->tv_sec && now.tv_nsec > abstime->tv_nsec))
+
-		   break;
+   gettimeofday(&tm, NULL);
+   now.tv_sec = tm.tv_sec;
+   now.tv_nsec = tm.tv_usec * 1000;
+   s64 timeout = (abstime->tv_sec - now.tv_sec) * 1000000000 + (abstime->tv_nsec - now.tv_nsec);
+
+   if (timeout >= 0) {
+      LightLock_Unlock(mutex);
+      LightLock_Unlock(&cond_data->lock);
+      if (svcWaitSynchronization(cond_data->semaphore, timeout))
+         retval = ETIMEDOUT;
+
+      LightLock_Lock(mutex);
+   } else {
+      retval = ETIMEDOUT;
    }
 
-   return 0;
+   return retval;
 }
 
 static INLINE int pthread_cond_init(pthread_cond_t *cond,
       const pthread_condattr_t *attr)
 {
-   LightEvent_Init((LightEvent *)cond, RESET_ONESHOT);
+   cond_t *cond_data = calloc(1, sizeof(cond_t));
+   if (!cond_data)
+      goto error;
+
+   if (svcCreateSemaphore(&cond_data->semaphore, 0, 1))
+      goto error;
+
+   LightLock_Init(&cond_data->lock);
+   cond_data->waiting = 0;
+   *cond = (pthread_cond_t)cond_data;
    return 0;
+
+ error:
+   svcCloseHandle(cond_data->semaphore);
+   if (cond_data)
+      free(cond_data);
+   return -1;
 }
 
 static INLINE int pthread_cond_signal(pthread_cond_t *cond)
 {
-   LightEvent_Signal((LightEvent *)cond);
+   int32_t count;
+   cond_t *cond_data = (cond_t *)*cond;
+   LightLock_Lock(&cond_data->lock);
+   if (cond_data->waiting) {
+      cond_data->waiting--;
+      LightLock_Unlock(&cond_data->lock);
+      svcReleaseSemaphore(&count, cond_data->semaphore, 1);
+   } else {
+      LightLock_Unlock(&cond_data->lock);
+   }
    return 0;
 }
 
 static INLINE int pthread_cond_broadcast(pthread_cond_t *cond)
 {
-   LightEvent_Signal((LightEvent *)cond);
+   int32_t count;
+   cond_t *cond_data = (cond_t *)*cond;
+   LightLock_Lock(&cond_data->lock);
+   while (cond_data->waiting) {
+      cond_data->waiting--;
+      LightLock_Unlock(&cond_data->lock);
+      svcReleaseSemaphore(&count, cond_data->semaphore, 1);
+      LightLock_Lock(&cond_data->lock);
+   }
+   LightLock_Unlock(&cond_data->lock);
+
    return 0;
 }
 
 static INLINE int pthread_cond_destroy(pthread_cond_t *cond)
 {
-   /*nothing to destroy*/
+   if (*cond) {
+      cond_t *cond_data = (cond_t *)*cond;
+
+      svcCloseHandle(cond_data->semaphore);
+      free(cond_data);
+   }
+   *cond = 0;
    return 0;
 }
 

retroarch.c

@@ -375,6 +375,7 @@ static const audio_driver_t *audio_drivers[] = {
 #ifdef _3DS
    &audio_ctr_csnd,
    &audio_ctr_dsp,
+   &audio_ctr_dsp_thread,
 #endif
 #ifdef SWITCH
    &audio_switch,
@@ -38459,6 +38460,13 @@ static enum runloop_state runloop_check_state(
       return RUNLOOP_STATE_QUIT;
 #endif
 
+#ifdef _3DS
+   /* Should be called once per frame */
+   if (!aptMainLoop())
+      return RUNLOOP_STATE_QUIT;
+#endif
+
+
    BIT256_CLEAR_ALL_PTR(&current_bits);
 
    p_rarch->input_driver_block_libretro_input   = false;

retroarch.h

@@ -605,6 +605,7 @@ extern audio_driver_t audio_psp;
 extern audio_driver_t audio_ps2;
 extern audio_driver_t audio_ctr_csnd;
 extern audio_driver_t audio_ctr_dsp;
+extern audio_driver_t audio_ctr_dsp_thread;
 extern audio_driver_t audio_switch;
 extern audio_driver_t audio_switch_thread;
 extern audio_driver_t audio_switch_libnx_audren;