PipeWire: Optimization to multi-threaded mode.

extras/backends/pipewire/miniaudio_pipewire.c

@@ -1355,7 +1355,7 @@ typedef struct
     ma_uint32 bufferSizeInFrames;
     ma_uint32 bufferCount;
     ma_uint32 rbSizeInFrames;
-    ma_pcm_rb rb;               /* For playback, PipeWire will read from this ring buffer. For capture, it'll write to it. */
+    ma_pcm_rb rb;                       /* Only used in single-threaded mode. Acts as an intermediary buffer for step(). For playback, PipeWire will read from this ring buffer. For capture, it'll write to it. */
     ma_device_descriptor* pDescriptor;  /* This is only used for setting up internal format. It's needed here because it looks like the only way to get the internal format is via a stupid callback. Will be set to NULL after initialization of the PipeWire stream. */
 } ma_pipewire_stream_state;
 
@@ -2353,11 +2353,16 @@ static void ma_stream_event_process__pipewire(void* pUserData, ma_device_type de
         pContextStatePipeWire->pw_stream_get_time_n(pStreamState->pStream, &time, sizeof(time));
 
         if (pStreamState->rbSizeInFrames > 0) {
+            if (ma_device_get_threading_mode(pDeviceStatePipeWire->pDevice) == MA_THREADING_MODE_SINGLE_THREADED) {
                 ma_pcm_rb_uninit(&pStreamState->rb);
             }
+        }
 
         pStreamState->rbSizeInFrames = (ma_uint32)time.size;
+
+        if (ma_device_get_threading_mode(pDeviceStatePipeWire->pDevice) == MA_THREADING_MODE_SINGLE_THREADED) {
             ma_pcm_rb_init(pStreamState->format, pStreamState->channels, pStreamState->rbSizeInFrames, NULL, ma_device_get_allocation_callbacks(pDeviceStatePipeWire->pDevice), &pStreamState->rb);
+        }
 
         pStreamState->initStatus |= MA_PIPEWIRE_INIT_STATUS_HAS_LATENCY;
         return;
@@ -2383,9 +2388,11 @@ static void ma_stream_event_process__pipewire(void* pUserData, ma_device_type de
     MA_PIPEWIRE_ASSERT(pBuffer->buffer->datas[0].data != NULL);
 
     if (frameCount > 0) {
-        ma_uint32 framesRemaining = frameCount;
-
         if (deviceType == ma_device_type_playback) {
+            if (ma_device_get_threading_mode(pDeviceStatePipeWire->pDevice) == MA_THREADING_MODE_MULTI_THREADED) {
+                ma_device_handle_backend_data_callback(pDeviceStatePipeWire->pDevice, pBuffer->buffer->datas[0].data, NULL, frameCount);
+            } else {
+                ma_uint32 framesRemaining = frameCount;
                 ma_uint32 framesAvailable = ma_pcm_rb_available_read(&pStreamState->rb);
 
                 /* Copy data in. Read from the ring buffer, output to the PipeWire buffer. */
@@ -2393,6 +2400,11 @@ static void ma_stream_event_process__pipewire(void* pUserData, ma_device_type de
                     /* Underflow. Just write silence. */
                     MA_PIPEWIRE_ZERO_MEMORY((char*)pBuffer->buffer->datas[0].data + (frameCount - framesRemaining) * bytesPerFrame, framesRemaining * bytesPerFrame);
                 } else {
+                    /*
+                    Two ways of handling this. In multi-threaded mode, it doesn't really matter which thread does the data
+                    processing so we can just do it directly from here and bypass the ring buffer entirely.
+                    */
+                    
                     while (framesRemaining > 0) {
                         ma_uint32 framesToProcess = (ma_uint32)ma_min(framesRemaining, framesAvailable);
                         void* pMappedBuffer;
@@ -2413,9 +2425,14 @@ static void ma_stream_event_process__pipewire(void* pUserData, ma_device_type de
                         }
                     }
                 }
+            }
 
             /*printf("(Playback) Processing %d frames... %d %d\n", (int)frameCount, (int)pBuffer->requested, pBuffer->buffer->datas[0].maxsize / bytesPerFrame);*/
         } else {
+            if (ma_device_get_threading_mode(pDeviceStatePipeWire->pDevice) == MA_THREADING_MODE_MULTI_THREADED) {
+                ma_device_handle_backend_data_callback(pDeviceStatePipeWire->pDevice, NULL, pBuffer->buffer->datas[0].data, frameCount);
+            } else {
+                ma_uint32 framesRemaining = frameCount;
                 ma_uint32 framesAvailable = ma_pcm_rb_available_write(&pStreamState->rb);
 
                 /* Copy data out. Write from the PipeWire buffer to the ring buffer. */
@@ -2438,6 +2455,7 @@ static void ma_stream_event_process__pipewire(void* pUserData, ma_device_type de
                         break;
                     }
                 }
+            }
 
             /*printf("(Capture) Processing %d frames...\n", (int)frameCount);*/
         }
@@ -2722,6 +2740,7 @@ static void ma_device_uninit__pipewire(ma_device* pDevice)
 {
     ma_device_state_pipewire* pDeviceStatePipeWire = ma_device_get_backend_state__pipewire(pDevice);
     ma_context_state_pipewire* pContextStatePipeWire = ma_context_get_backend_state__pipewire(ma_device_get_context(pDevice));
+    ma_device_type deviceType = ma_device_get_type(pDevice);
 
     if (pDeviceStatePipeWire->capture.pStream != NULL) {
         pContextStatePipeWire->pw_stream_destroy(pDeviceStatePipeWire->capture.pStream);
@@ -2740,6 +2759,15 @@ static void ma_device_uninit__pipewire(ma_device* pDevice)
     pContextStatePipeWire->pw_context_destroy(pDeviceStatePipeWire->pContext);
     pContextStatePipeWire->pw_loop_destroy(pDeviceStatePipeWire->pLoop);
 
+    if (ma_device_get_threading_mode(pDeviceStatePipeWire->pDevice) == MA_THREADING_MODE_SINGLE_THREADED) {
+        if (deviceType == ma_device_type_capture || deviceType == ma_device_type_duplex) {
+            ma_pcm_rb_uninit(&pDeviceStatePipeWire->capture.rb);
+        }
+        if (deviceType == ma_device_type_playback || deviceType == ma_device_type_duplex) {
+            ma_pcm_rb_uninit(&pDeviceStatePipeWire->playback.rb);
+        }
+    }
+
     ma_free(pDeviceStatePipeWire, ma_device_get_allocation_callbacks(pDevice));
 }
 
@@ -2801,6 +2829,8 @@ static ma_result ma_device_step__pipewire(ma_device* pDevice, ma_blocking_mode b
             return MA_DEVICE_NOT_STARTED;
         }
 
+        /* We need only process data here in single-threaded mode. */
+        if (ma_device_get_threading_mode(pDeviceStatePipeWire->pDevice) == MA_THREADING_MODE_SINGLE_THREADED) {
             /* We want to handle both playback and capture in a single iteration for duplex mode. */
             if (deviceType == ma_device_type_capture || deviceType == ma_device_type_duplex) {
                 ma_uint32 framesAvailable;
@@ -2847,6 +2877,7 @@ static ma_result ma_device_step__pipewire(ma_device* pDevice, ma_blocking_mode b
                     }
                 }
             }
+        }
 
         if (hasProcessedData || blockingMode == MA_BLOCKING_MODE_NON_BLOCKING) {
             break;